Dados do Trabalho

Título

IS IT POSSIBLE TO ASSESS INTRINSIC AND ACQUIRED RESISTANCE TO NEOADJUVANT CHEMOTHERAPY USING A NOVEL IN VITRO BREAST CANCER CHEMORESISTANCE PLATFORM?

Trabalho aprovado na Plataforma Brasil? Inserir número do CAAE.

59609822.4.0000.5530

Introdução

Tumor resistance is the main cause of treatment failure leading to cancer progression and is classified into intrinsic (preexisting condition) and acquired resistance (induced by a drug). Some methods are available worldwide to assess drug resistance, however, no in vitro chemoresistance test is currently approved for clinical use in Brazil.

Justificativa

Individualized treatment based on personal tumor characteristics is essential for reducing toxicity, improving therapeutic outcomes and patient survival.

Objetivo

Our preliminary study aims to validate the efficacy of a novel in vitro chemoresistance platform to demonstrate intrinsic and acquired resistance in breast cancer undergoing neoadjuvant chemotherapy (NACT). We also evaluated the Gene Expression Omnibus (GEO) using a machine learning technique to classify genes involved in NACT resistance.

Métodos

Patients with invasive BC who underwent biopsy and who presented residual disease after NACT were included. Fresh tumor samples were collected during biopsy or surgery and dissociated to obtain the tumor cells. The tumor cells were cultured in the chemoresistance platform with cytotoxic drugs (doxorubicin, epirubicin, paclitaxel, docetaxel, cyclophosphamide; carboplatin and capecitabine in selected cases), and after 72h cell viability was evaluated. The test result is defined based on cell viability as low (< 40%), medium (40-60%), and high (> 60%) resistance. In addition, we used data from the GEO portal (GSE25066 project) to analyze the gene expression patterns and outcomes of breast cancer patients undergoing NACT. We selected 488 samples and divided into complete pathological response (pCR) (n = 99) and residual disease (RD) (n = 389) groups. Following data processing we obtained a comprehensive set of 12,644 genes. To classify the samples into pCR and RD categories, we implemented the XGBoost algorithm, a machine learning technique based on trees. We used the SHAP (SHAPley Additive explanations) method to interpret prediction results of the machine learning model.

Resultados

Samples from 15 biopsies (primary tumors) and 13 RD after NACT were included. Samples collected during biopsies showed a higher prevalence of Luminal A (46.6%), followed by Luminal B (20%), Lum B/HER2+ (13.4%), triple-negative breast cancer (TNBC) (13.4%) and HER2+ (6.6%). These samples displayed increased rates of high resistance to docetaxel (73%) and paclitaxel (73%) compared with doxorubicin (13.3%), epirubicin (20%), and cyclophosphamide (6.6%). Five patients were referred to NACT. Among them, two patients diagnosed with Luminal B and TNBC did not respond to treatment with carboplatin and docetaxel, and the chemoresistance platform confirmed a high resistance to both drugs. The remaining three patients (one Luminal A, one HER2, and one TNBC) are currently undergoing AC-T (doxorubicin plus cyclophosphamide followed by paclitaxel) treatment, having completed the AC regimen and presenting a good clinical response. In the chemoresistance platform, none of the tumors displayed high resistance to doxorubicin and cyclophosphamide. Regarding molecular subtypes of residual tumors after NACT: five were TNBC (38.5%), four Luminal (30.7%), and four Luminal/HER2+ (30.8%). Most (69.3%) of the patients received ACT + Carboplatin, and 30.7% received docetaxel + carboplatin + trastuzumab + pertuzumab. The chemoresistance platform revealed that residual tumors after NACT had higher rates of drugs resistance: 100% showed high resistance to docetaxel, 92.3% to paclitaxel, 38.4% to doxorubicin, 41.6% to epirubicin, 15.4% to cyclophosphamide and 80% to carboplatin. During a median follow-up of 5 months, three patients experienced disease progression while using adjuvant capecitabine, and in the chemoresistance platform, all of them presented high resistance to the drug. The XGBoost algorithm achieved an average accuracy of 78% and average precision of 62% in classifying samples into pCR and RD. Through the SHAP method, 163 genes were selected, and the top five genes that contributed the most to the model were RPN2, DCLK2, PELP1, TUBBA2A and CD24. These findings underscore the significance of these genes in predicting the response to breast cancer treatment.  

Conclusão

Our preliminary findings highlighted the efficacy of the in vitro chemoresistance platform to demonstrate the involvement of intrinsic resistance in tumors that did not exhibit a pCR to NACT and revealed a change in resistance profile after NACT, which could potentially contribute to the worse prognosis of these patients associated with the acquisition of resistance. Our machine learning model supports the hypothesis that an intrinsic resistance is present in patients who do not achieve a pCR to NACT and suggests possible genes involved in tumor resistance.