Immunogenomic profiling and pathological response results from a clinical trial of docetaxel and carboplatin in triple-negative breast cancer

Patients with triple-negative breast cancer (TNBC) who do not achieve pathological complete response (pCR) following neoadjuvant chemotherapy have a high risk of recurrence and death. Molecular characterization may identify patients unlikely to achieve pCR. This neoadjuvant trial was conducted to determine the pCR rate with docetaxel and carboplatin and to identify molecular alterations and/or immune gene signatures predicting pCR. Patients with clinical stages II/III TNBC received 6 cycles of docetaxel and carboplatin. The primary objective was to determine if neoadjuvant docetaxel and carboplatin would increase the pCR rate in TNBC compared to historical expectations. We performed whole-exome sequencing (WES) and immune profiling on pre-treatment tumor samples to identify alterations that may predict pCR. Thirteen matching on-treatment samples were also analyzed to assess changes in molecular profiles. Fifty-eight of 127 (45.7%) patients achieved pCR. There was a non-significant trend toward higher mutation burden for patients with residual cancer burden (RCB) 0/I versus RCB II/III (median 80 versus 68 variants, p 0.88). TP53 was the most frequently mutated gene, observed in 85.7% of tumors. EGFR, RB1, RAD51AP2, SDK2, L1CAM, KPRP, PCDHA1, CACNA1S, CFAP58, COL22A1, and COL4A5 mutations were observed almost exclusively in pre-treatment samples from patients who achieved pCR. Seven mutations in PCDHA1 were observed in pre-treatment samples from patients who did not achieve pCR. Several immune gene signatures including IDO1, PD-L1, interferon gamma signaling, CTLA4, cytotoxicity, tumor inflammation signature, inflammatory chemokines, cytotoxic cells, lymphoid, PD-L2, exhausted CD8, Tregs, and immunoproteasome were upregulated in pre-treatment samples from patients who achieved pCR. Neoadjuvant docetaxel and carboplatin resulted in a pCR of 45.7%. WES and immune profiling differentiated patients with and without pCR. Trial registration: Clinical trial information: NCT02124902, Registered 24 April 2014 & NCT02547987, Registered 10 September 2015.


Introduction
Triple-negative breast cancer (TNBC) is a heterogeneous clinical breast cancer subtype characterized by the absence of expression of receptors for estrogen (ER), progesterone (PR), and the lack of overexpression of the tyrosine kinase cell surface receptor HER2/Neu (HER2). This disease subset accounts for approximately 15-20% of all patients with primary breast cancer [1]. TNBC is characterized by a poorer prognosis compared to other clinical subtypes, and because of the absence of targetable receptors, chemotherapy remains the principal systemic therapy.
Neoadjuvant chemotherapy is widely used in early-stage patients with TNBC who are being treated with curative intent. The preoperative approach provides an opportunity to assess in vivo responses, enables rapid identification of effective drugs, and allows tailoring of adjuvant systemic therapy [2]. Due to the improvements in mortality, anthracycline-and taxane-based regimens are widely used in the neoadjuvant treatment of patients with TNBC [3]. Unfortunately, anthracyclines are associated with increased cardiac mortality, myelodysplastic syndromes, and treatment-related leukemia [3,4]. Thus, there is interest in evaluating other effective regimens for patients with TNBC. Platinum salts induce double-strand DNA damage and are active in BRCA -associated breast cancers [5,6]. Sporadic TNBC and BRCA -associated breast cancers share similar molecular features [7], suggesting possible benefits of platinum salts in patients with TNBC. Recent studies have shown that platinum plus taxane-based non-anthracycline regimens may be an alternative in TNBC patients [8][9][10][11].
The failure of chemotherapy to eradicate disease is believed to be due to selection of cells intrinsically resistant to chemotherapy [12]. Only 30-50% of patients with TNBC who receive neoadjuvant chemotherapy achieve a pathological complete response (pCR) [13][14][15][16]. Patients who do not achieve a pCR tend to have a higher rate of recurrence and poorer overall survival than patients who do achieve pCR [17][18][19][20]. The 3-year risk of distant recurrence for non-pCR patients is 27% versus 9% for those achieving pCR [21]. Three-year survival probability is only 68% in non-pCR versus 94% in TNBC patients who achieve pCR [17]. The median survival once TNBC has recurred is only 13-25 months [22][23][24].
There is no reliable method for predicting which individual patient will achieve pCR; consequently, many earlystage patients with TNBC may be exposed to several months of ineffective chemotherapy prior to definitive breast surgery. The ability to individualize chemotherapy is yet to be achieved because robust predictive markers for chemotherapy response have not been identified. We therefore sought to determine the pCR rate with a non-anthracycline regimen of docetaxel and carboplatin in TNBC. Using this clinical trial as a platform for biomarker discovery, we also sought to understand the role of the host and tumor immune profile and tumor genomics in pathological responses.

Patient population
Eligible patients included pre-or post-menopausal women at least 18 years old, with clinical stages II or III ER negative (Allred score < 3 or less than 1% positive staining cells in the invasive component of the tumor) and HER2 negative (0 or 1 + by IHC or FISH negative) invasive breast cancer. Additional eligibility criteria include Eastern Cooperative Oncology Group Performance Status of 0-2, adequate organ and marrow function, tumor size ≥ 2 cm in one dimension by clinical or radiographic exam (World Health Organization criteria), and patients with palpable lymph nodes regardless of tumor size. Exclusion criteria included prior treatment of the current cancer, uncontrolled intercurrent illness, bilateral or inflammatory cancer, pregnant/nursing, and prior sentinel lymph node biopsy.

Study procedures
All patients were treated with neoadjuvant intravenous docetaxel 75 mg/m 2 and carboplatin AUC 6 cycled every 21 days for 6 cycles, with granulocyte colony-stimulating factor support. Dose adjustments for toxicity were at the discretion of the treating physician. Definitive surgery was performed 3-5 weeks after completion of chemotherapy. Patients received adjuvant radiation when indicated, and adjuvant chemotherapy for patients without pCR was left to the discretion of the treating physician. NCI CTCAE 4.0 was used to record severity and attribution of toxicities. Research tumor biopsies for correlative studies were obtained at baseline prior to chemotherapy, on cycle 1 day 3 (C1D3), and at time of definitive surgery following neoadjuvant chemotherapy in those patients with residual disease. On-treatment biopsy on C1D3 and biopsy at time of relapse were optional. Figure 1 shows the trial schema.

Residual cancer burden (RCB) scoring
Histologic slides from surgical cases post-neoadjuvant therapy were reviewed by a breast pathologist (IH) to determine tumor bed size, percent neoplastic cellularity within the tumor bed, percent of residual tumor that was in situ, number of positive nodes, and largest lymph node deposit. RCB score and category were assigned by the MD Anderson method using a publicly available web calculator (http:// www3. mdand erson. org/ app/ medca lc/ index. cfm? pagen ame= jscon vert3, accessed 9/9/20).

Tumor-infiltrating lymphocytes (TILs) assessment
TILs were quantitated according to the method recommended by the International Immuno-Oncology Biomarker Working Group on Breast Cancer [25]. A breast pathologist (IH) independently reviewed whole slide scans (blinded to treatment status) and documented the percentage of TIL infiltration in increments of 5% in tumor-adjacent stroma.

Whole-exome sequencing (WES)
Tumor DNA was extracted from fresh-frozen biopsies and matched leukocyte germline DNA from blood samples. WES libraries were enriched using the NimbleGen Roche VCRome v2.1 hybrid capture reagent supplemented with a "panel killer" spike-in designed by BCM. Pairedend (2 × 150 base pair) next generation sequencing was performed using the Illumina platform to a target depth of coverage of 100 × (mean target coverage achieved was 70-130 × across the cohort). Sequence data processing pipelines utilized tools implemented in Docker containers (published to DockerHub), input parameters and data specified in YAML configuration files, pipeline steps and dependencies expressed using the Common Workflow Language (CWL), and compute tasks submitted to a compute cluster using Cromwell and platform LSF [26]. All associated CWL files, example YAML files, and docker files are version tracked using Git and GitHub (https:// github. com/ genome/ analy sisworkfl ows). Metadata on analysis runs was tracked using an Analysis Information Management System developed at the Washington University's McDonnell Genome Institute [27]. Briefly, WES analysis was performed by aligning sequence reads to the human reference genome build GRCh38 using the BWA-MEM aligner [28]. Alignments were subjected to base quality score recalibration [29], sorted by chromosome position, duplicates marked with Picard (http:// broad insti tute. github. io/ picard/), and converted to a lossless CRAM format (https:// github. com/ samto ols/ hts-specs) to reduce disk usage. All samples in each cohort were tested for sample swaps and contamination using Somalier (https:// github. com/ brentp/ somal ier) and subjected to a QC analysis consisting of Picard, samtools flagstat [30], and VerifyB-amID (https:// github. com/ statg en/ verif yBamID). Somatic single-nucleotide variants (SNVs) and small insertions and deletions were called using an ensemble approach involving Mutect2, Strelka2, VarScan2, and Pindel [31][32][33]. Somatic variants from each of these callers were left aligned and trimmed using GATK LeftAlignAndTrim [29], merged into a single variant call format file using GATK CombineVariants and multiallelic sites separated using vt decompose [34]. Somatic variants were subjected to a false-positive filter to flag variants of high frequency in the population according to GNOMAD [35], lacking minimum sequence read support from both sequence strands, corresponding to regions of the genome with ambiguous read mapping, and those that fail a log likelihood test that models the observed read support relative to known error rates of the sequencing platform and conservative sample specific assumptions of tumor purity. The resulting candidate somatic variants were subjected to a formal manual review standard operating procedure [36]. The resulting high confidence somatic variants were annotated for transcript variant effect using VEP and Ensembl transcripts [37,38]. Variant allele frequencies were computed using bam-readcount (https:// github. com/ genome/ bam-readc ount). Pathway analyses for genes harboring subclonal variants selected under treatment pressure were performed using WEB-based GEne SeT AnaLysis Toolkit [39] using an Over-Representation Analysis (ORA) approach and "Wikipathway cancer" as the source of pathways.

NanoString gene expression analysis
RNA extracted from formalin-fixed paraffin-embedded (FFPE) tissue samples was analyzed on the nCounter® analysis system using the PanCancer IO 360™ panel (for research use only). Raw data counts were normalized using the geomean of 20 housekeeping genes in the IO360 panel and each gene was adjusted based on IO360 panel standards to adjust for batch-to-batch variation. The housekeepernormalized and panel standard-normalized data are Log(2) transformed. IO360 gene analysis for 48 signatures measuring immune cell abundance, immune signaling, tumor, and stromal biology was calculated as previously described [40,41]. A constant of 8 is added to the tumor inflammation signature (TIS) so that it is on the same scale as investigational use only (IUO) TIS, making scores comparable across research use only (RUO) and IUO assays. Other non-TIS signatures are also adjusted with constants to express values in a similar range.

Statistical analyses
The primary endpoint of the clinical trial is pCR rate calculated as the percentage of patients who achieve pCR among all evaluable patients. pCR is defined as absence of residual invasive disease in the breast and lymph nodes following neoadjuvant chemotherapy. Exploratory aims were to investigate immune and genomic changes with the intent to identify predictors of response. A sample size of 100 patients provides 82.1% power to test a pCR rate of 40% against the null rate of 28% with standard chemotherapy, based on 1-sided binomial exact test at a target 0.05 alpha level. If 36 or more patients achieve a pCR, we conclude that the investigational regimen yields better efficacy than standard chemotherapy.
Patient characteristics were summarized by descriptive statistics, counts and percentages for categorical characteristics, and median with inter-quartile ranges (IQR) for quantitative characteristics. Tumor burden was compared between pCR (RCB 0) versus non-pCR patients, and between RCB 0/I versus RCB II/III patients, by Wilcoxon rank sum test and compared between paired pre-treatment and C1D3 samples by Wilcoxon signed-rank test. Mutation landscape waterfall plot was generated using the R package "GenVisR" (Version 1.16.1). Gene sets defined in the Molecular Signatures Database v7.1 (MSigDB v7.1 released on March 2020) were downloaded and extracted using R package "msigdbr." The overall effect of gene mutations of a gene set on pathological outcome was evaluated using the sequence kernel-based association (SKAT) test method in the logistic regression framework. SKAT p value was reported for each gene set.
NanoString IO360 signatures were compared based on pathological response and time point. Differential expression based on response was fit on a per-gene or per-signature basis using a linear model for analyses without a blocking factor. The statistical model uses the expression value or signature score as the dependent variable and fits a grouping variable as a fixed effect to test for differences in the levels of that grouping variable.
For differential expression for time series analysis, the duplicateCorrelation function within the limma R package is used to assess the correlation between subsequent time points. This correlation estimate is fit into the linear mixed effect model with subject as the random effect and the correlation between the repeated temporal measurements.
Expression (gene or signature) = + Response + For all differential expression analysis, all models are fit using the limma package in R.

Patient characteristics and clinical efficacy
Between 8/2014 and 1/2020, 168 patients were screened and 132 ultimately received protocol therapy. Thirty-five did not meet inclusion criteria/were not registered due to reasons such as being found to have estrogen receptorpositive breast cancer, metastatic disease, or abnormal laboratory values. Five patients were not evaluable for pCR due to a variety of reasons, including patient or physicians choice to withdraw protocol therapy. One hundred and twenty seven patients were evaluable. Median follow-up is 27  Expression_ (gene or signature) = + SubjectID + Group + Toxicity All patients who received at least 1 cycle of combination chemotherapy are evaluable for toxicity. Due to toxicities, 13 patients did not complete all 6 cycles of protocol-specified therapy (five had 4 cycles, and eight patients had 5 cycles of combination therapy). Treatment-emergent adverse events of any grade occurred in 123 patients, with 1947 events reported. Table 2 shows the incidence of grade 3 and 4 adverse events occurring in at least two patients. The incidence of grade 3-4 anemia was 18.2%, thrombocytopenia was 13.6%, diarrhea 9.1%, and febrile neutropenia 7.6%. Adverse events leading to discontinuation of the regimen were reported in 15 patients (11.4%). Nineteen patients (14.4%) had at least one SAE ( Table 2). No treatment-related deaths occurred.
Patients with higher post-treatment absolute lymphocyte counts (ALCs) from peripheral blood were more likely to achieve pCR than those with lower ALCs (OR 5.5; 95% CI 1.5-20.7, p = 0.011). Post-treatment median ALCs were 1.55 cells/mm 3 (range 0.8-3.6 cells/mm 3 ) in pCR patients and 1.4 cells/mm 3 (range 0.4-3.9 cells/mm 3 ) in those with residual disease. Similarly, patients with higher minimum ALCs were also more likely to achieve pCR than those with a lower ALC nadir (OR 9.1; 95% CI 1.5-54.9, p = 0.016).
The minimum ALC is defined as the ALC nadir that patients experienced during the period that they received neoadjuvant chemotherapy. The associations of post-treatment and minimum ALCs with pCR remained significant after adjusting for age and clinical stage at diagnosis (post-treatment ALC OR 7.6; 95% CI 1. (p = 0.028), were upregulated in pCR versus non-pCR samples (Fig. 2). Key immune signatures of borderline significance included T cells, CD8 T cells, IL7R, and TIGIT. A heatmap displayed as Additional file 2 uses unsupervised hierarchical clustering to show relatedness among signature scores for baseline samples according to pathological response.

Tumor mutation profiling and pathological response
WES was performed on baseline pre-treatment samples collected from a subset (N = 56 patients) of the clinical trial population. Due to funding limitations, only samples with the highest tumor cellularity for sequencing were selected for WES. Thirteen patients had baseline-matched samples from C1D3 that were analyzed. 9063 variants were detected in 5386 unique genes. The mutation landscape waterfall plot of genes with a mutation frequency of > 5% is shown in Fig. 4. The overall mutation burden for patients who achieved pCR was not significantly different from non-pCR patients (median of 78.5 variants, IQR 43-134 in pCR, vs median 72, IQR 47.8-103.8 in non-pCR, Wilcoxon rank sum test p = 0.98). Similarly, there was no difference in the overall mutation burden for patients with RCB 0/I versus those with RCB II/III (median of 80 variants, IQR 40-134 in RCB 0/I, vs median 68, IQR 53.5-87.8 in RCB II/III, Wilcoxon rank sum test p = 0.88) (Fig. 5). Table 3 shows the genes with variants occurring in at least 10% of the biomarker population, according to pathological response. TP53 was the most frequently mutated gene observed in 48 of 56 patients sequenced (85.7%). As expected for TP53, we observed mutations primarily within the DNA-binding domain, and many of the mutations are likely loss-of-function variants caused One of the patients with an EGFR mutation has recurred, despite pCR. In the longitudinal analyses, there was evidence of tumor heterogeneity and shifts in clonal architecture under treatment pressure from pre-treatment to C1D3, due to both selection and depletion of subclones. However, more variants were enriched suggesting possible subclone emergence under treatment pressure (Fig. 6). The overall variant counts in the matched samples at C1D3 trended higher (median of 82, IQR 49-157) than corresponding pre-treatment samples (median of 72, IQR 42-92), p = 0.29. As expected, all tumor pairs had a substantial set of shared clonal variants with TP53 variants exhibiting high variant allele frequency (VAF) reflecting its driver status. Across all patients, 289 genes harbored such emerging variants including a second TP53 Y220C variant in patient NTN022 (0% VAF at baseline and 13.0% VAF at C1D3) and an EGFR S768I variant in NTN046 (0.07% VAF at baseline and 29.5% VAF at C1D3). Pathway analysis of the entire set of mutated genes emerging at C1D3 (green points in Fig. 6) showed significant enrichment for DNA damage and nucleotide synthesis pathways suggesting possible selection for cells with molecular mechanisms of resistance to the DNA damaging agent carboplatin.
Last, we explored whether mutations in different gene families would cluster in recognized pathways. Using the Molecular Signatures Database (MsigDB) v7.1, several gene families involved in immune signature-related gene sets (MsigDB C7 set) showed differences between RCB 0/I and RCB II/III samples (Additional file 6). Additionally, borderline differences in PI3K AKT MTOR signaling pathway among MsigDB 50 hallmark gene sets were identified between RCB 0/I and RCB II/III samples. There were no differences in inflammatory response, angiogenesis, apoptosis, NOTCH signaling, TNF alpha, or androgen response pathways.

Discussion
In this multicenter single-arm phase II trial of neoadjuvant docetaxel and carboplatin in patients with newly diagnosed clinical stage II-III TNBC, we observed a pCR rate of 45.7%. These data are consistent with previous observations that non-anthracycline-based regimens in TNBC patients achieve similar pCR rates as anthracycline plus taxane-based regimens [8][9][10][11]. Recent randomized trials demonstrate that the addition of carboplatin to anthracycline taxane-based regimens not only increase pCR rates but also increase toxicity [13,14]. More recently, Sharma et al. compared neoadjuvant carboplatin with docetaxel to carboplatin with paclitaxel followed by anthracyclines in patients with earlystage TNBC [42]. This study demonstrated similar pCR rates in both study groups: carboplatin with docetaxel group pCR 52% versus anthracycline group pCR 55%, p = 0.84. Patients who did not receive anthracyclines had a more favorable toxicity profile and higher treatment completion rate compared with patients who received anthracyclines. The effects of the addition of carboplatin to anthracycline-based regimens on longer-term clinical outcomes such as event-free and overall survival are conflicting [43]. GeparSixto, a randomized phase 3 trial that evaluated the addition of carboplatin to anthracycline and taxanes in TNBC, showed a higher 3-year disease-free survival rate in the carboplatin group Fig. 3 Differential expression of IO360 gene signatures before and on-treatment (N = 66 patients). Volcano plot showing differential expression of IO360 immune signatures from baseline or on treatment samples. Signatures in blue were significantly different using an adjusted p-value (p < 0.05). Dotted lines indicate an unadjusted p-value of p < 0.05 and p < 0.01. Larger circles indicate greater significance vs the non-carboplatin group (85% vs 76%, p = 0.03) [14,44]. Conversely, Cancer and Leukemia Group B (CALGB) 40603 randomized phase II trial demonstrated no difference in 3-year event-free survival with the addition of carboplatin to anthracycline taxane-based neoadjuvant therapy (71% versus 76%, p = 0.36) [45]. As a result, carboplatin has not been incorporated into the routine clinical management of early-stage TNBC. Ongoing clinical trials may address long-term outcomes in the future [46][47][48]. The similarity in antitumor activity with docetaxel and carboplatin compared with standard anthracycline taxane-based regimens, and the increased toxicities with adding carboplatin to anthracycline taxane-based regimens provide rationale for considering docetaxel and carboplatin chemotherapy for patients with early-stage TNBC. Moreover, the long-term risk of cardiotoxicity, myelodysplasia, and therapy-associated leukemia is minimized with non-anthracycline regimens.
pCR is a surrogate of long-term outcomes in patients with TNBC, and the extent of residual disease is linked to the risk of recurrence [16,17,[49][50][51]. Individuals who do not achieve pCR or RCB I have a high risk of recurrent disease and subsequent death, with a hazard ratio for an overall survival event reported to be as high as 12.4 (95% CI 5.8-26.5, p = 0.001) [16,17]. More recently, results from the ECOG-ACRIN EA 1131 showed that patients who did not achieve pCR had an extremely poor 3-year invasive disease-free Fig. 4 The mutation landscape waterfall plot of genes with a nonsynonymous (frameshift, in-frame deletion, in-frame insertion, mature miRNA, missense, protein altering, splice acceptor, splice donor, splice region, start lost, stop gained, and stop lost) mutation frequency > 5%. Top bar plot indicates mutation burden, left bar plot indicates mutation frequency, and lower panel provides clinical annotation survival of less than 50% [52]. Other than BRCA germline status [44,53,54], there are no predictive factors for pCR. There are promising data using early imaging changes in positron emission or computed tomography, tumor morphological changes, and ctDNA to predict pathological response to neoadjuvant chemotherapy [55][56][57][58]. From a clinical perspective, understanding variables that predict response is an urgent unmet need. If TNBC patients who will not achieve pCR can be identified earlier, they may be triaged to innovative trials or definitive surgery, with a view to changing the natural history of resistant TNBC while sparing them the toxicity of ineffective chemotherapy. Therefore, we examined tumor immune and genomic profiles to identify molecular factors that may predict response. Our study did not identify mutation burden as a predictor for pathological response to docetaxel and carboplatin chemotherapy. An interesting observation was the identification of EGFR mutations in pCR patients. Although EGFR mutations have previously been described in TNBC [59], the clinical relevance has not been described. These results need to be confirmed in a larger study, but may suggest a potential rationale for evaluating EGFR tyrosine kinase inhibitor therapy in TNBC patients harboring EGFR mutations.
Due to the absence of recurrently mutated genes other than TP53 [60], the aggregation of individual genes at the pathway level may be a more practical way to evaluate predictors of response in TNBC. On-treatment samples had a suggestion of clone emergence, with increased variants compared to baseline samples. Pathway analysis of the emerging genes showed enrichment for DNA damage pathway, suggesting selection for resistance to the DNA damaging agents. These hypothesis-generating results suggest that early on-treatment tumor assessment may be used in the future to identify patients who may be more likely to respond to chemotherapy. We also found that several immune-related pathway gene signatures showed differences between pathological outcomes, suggesting the possible utility of immune perturbation in early TNBC. Patients with higher baseline TILs, baseline upregulation of immunerelated (IDO, PD-L1, lymphoid, Tregs), inflammationrelated (inflammatory cytokines, immunoproteasome), and cytotoxicity-related (interferon gamma signaling, cytotoxic cells, exhausted CD8) gene signatures were more likely to achieve pCR. In addition, patients who were able to maintain higher post-treatment absolute lymphocyte counts despite the lymphodepleting effects of chemotherapy were also more likely to achieve pCR. Taken together, these results suggest that patients with an inflamed tumor microenvironment  (TME) that experience less treatment-related lymphopenia may be more responsive to neoadjuvant chemotherapy with docetaxel and carboplatin. Thus, immunomodulatory interventions such as IL-7 treatment which is aimed at increasing the persistence, survival, and trafficking of lymphocytes may provide a significant clinical benefit when combined with neoadjuvant strategies. Long-acting recombinant human IL-7 (rhIL-7-hyFc) has shown to be safe and well tolerated in human, while significantly increasing the frequency of lymphocytes [61]. rhIL-7-hyFc has also proven to increase T cell infiltration and inflammation within the TME in cancerbearing mouse models [62]. Using rhIL-7-hyFc treatment to improve the rates of pCR in TNBC patients undergoing adjuvant chemotherapy is an appealing strategy that needs further investigation. Strengths of this translational study include the high proportion of African Americans accrued, which is likely reflective of the demographics of both accrual sites, and the commitment of the investigators to accrue minorities to clinical trials. Furthermore, this was a multicenter collaborative study, which simultaneously served as a platform for biomarker analyses. Additionally, this was a homogenously treated, clinically well-annotated TNBC cohort. All samples were processed similarly at the same institution, the whole-exome sequencing approach utilized was comprehensive, and we had normal tissue on all patients for accurate variant calling. However, several limitations deserve comment. First, this is a phase 2 single-arm nonrandomized study. Therefore, a larger randomized trial is necessary for further evaluation of non-anthracyclinebased regimens in TNBC. Second, the relatively small sample size of the correlative study limits the power of the analyses. Follow-up is short but ongoing, and longterm clinical outcomes will be reported as results become The clonal status of variants in each sample is indicated by color. Red represents subclonal depleted mutations, green represents subclonal selected mutations, and gray represents shared clonal mutations. TP53 variants were called in 11 of the participants' samples and are highlighted in the plots by the triangle available. Despite these limitations, this is the first hybrid study with comprehensive tumor profiling comparing a homogeneously treated cohort of TNBC patients to identify biomarkers predicting pCR.

Conclusion
Our results demonstrate a robust pCR rate of 45.7% with this alternative non-anthracycline chemotherapy regimen administered to patients with TNBC. Tumor mutation assessment for EGFR, RB1, RAD51AP2, SDK2, L1CAM, KPRP, PCDHA1, CACNA1S, CFAP58, COL22A1, COL4A5 mutations, and immune-related gene signatures may discriminate patients who will achieve pCR following platinum-based neoadjuvant chemotherapy. While larger confirmatory studies are needed, these may be potential biomarkers for predicting pathological response and therapeutic efficacy in patients with TNBC and emphasize the need for molecular analyses in therapeutic clinical trials.

Data availability
The datasets used and/or analyzed during the present study are available from the corresponding author on reasonable request.

Declarations
Conflict of interest FOA reports consulting for Eisai, Immunomedics, AstraZeneca, Athenex, Cardinal Health, Pfizer, AbbVie, Best Doctors, and Advance Medical. FOA reports contracted research for Immunomedics, Pfizer, Seattle Genetics, NeoImmuneTech, RNA Diagnostics, and Astellas. MFR reports consulting for Genentech, MacroGenics, Daiichi, Seattle Genetics, and Novartis. MFR reports contracted research for Pfizer. RB consulting for Genentech. RB reports contracted research for Puma Biotechnology, Inc.

Ethical Approval
The protocol and informed consent documents were approved by WUSM and BCM. Upon approval, all participating institutions agreed to follow the Declaration of Helsinki, good clinical practice guidelines, and the applicable parts of the U.S. Code of Federal Regulations.
Informed consent Written informed consent was required for enrollment.