A framework for validating AI in precision medicine: considerations from the European ITFoC consortium.

BACKGROUND: Artificial intelligence (AI) has the potential to transform our healthcare systems significantly. New AI technologies based on machine learning approaches should play a key role in clinical decision-making in the future. However, their implementation in health care settings remains limited, mostly due to a lack of robust validation procedures. There is a need to develop reliable assessment frameworks for the clinical validation of AI. We present here an approach for assessing AI for predicting treatment response in triple-negative breast cancer (TNBC), using real-world data and molecular -omics data from clinical data warehouses and biobanks. METHODS: The European "ITFoC (Information Technology for the Future Of Cancer)" consortium designed a framework for the clinical validation of AI technologies for predicting treatment response in oncology. RESULTS: This framework is based on seven key steps specifying: (1) the intended use of AI, (2) the target population, (3) the timing of AI evaluation, (4) the datasets used for evaluation, (5) the procedures used for ensuring data safety (including data quality, privacy and security), (6) the metrics used for measuring performance, and (7) the procedures used to ensure that the AI is explainable. This framework forms the basis of a validation platform that we are building for the "ITFoC Challenge". This community-wide competition will make it possible to assess and compare AI algorithms for predicting the response to TNBC treatments with external real-world datasets. CONCLUSIONS: The predictive performance and safety of AI technologies must be assessed in a robust, unbiased and transparent manner before their implementation in healthcare settings. We believe that the consideration of the ITFoC consortium will contribute to the safe transfer and implementation of AI in clinical settings, in the context of precision oncology and personalized care.

Overexpression of MLN51 triggers P-body disassembly and formation of a new type of RNA granules.

Metastatic lymph node 51 (MLN51, also known as CASC3) is a core component of the exon junction complex (EJC), which is loaded onto spliced mRNAs and plays an essential role in determining their fate. Unlike the three other EJC core components [eIF4AIII, Magoh and Y14 (also known as RBM8A)], MLN51 is mainly located in the cytoplasm, where it plays a key role in the assembly of stress granules. In this study, we further investigated the cytoplasmic role of MLN51. We show that MLN51 is a new component of processing bodies (P-bodies). When overexpressed, MLN51 localizes in novel small cytoplasmic foci. These contain RNA, show directed movements and are distinct from stress granules and P-bodies. The appearance of these foci correlates with the process of P-body disassembly. A similar reduction in P-body count is also observed in human HER2-positive (HER2(+)) breast cancer cells overexpressing MLN51. This suggests that P-body disassembly and subsequent mRNA deregulation might correlate with cancer progression.

COUP-TFI modifies CXCL12 and CXCR4 expression by activating EGF signaling and stimulates breast cancer cell migration.

BACKGROUND: The orphan receptors COUP-TF (chicken ovalbumin upstream promoter transcription factor) I and II are members of the nuclear receptor superfamily that play distinct and critical roles in vertebrate organogenesis. The involvement of COUP-TFs in cancer development has recently been suggested by several studies but remains poorly understood. METHODS: MCF-7 breast cancer cells overexpressing COUP-TFI and human breast tumors were used to investigate the role of COUP-TFI in the regulation of CXCL12/CXCR4 signaling axis in relation to cell growth and migration. We used Immunofluorescence, western-blot, RT-PCR, Formaldehyde-assisted Isolation of Regulatory Elements (FAIRE) assays, as well as cell proliferation and migration assays. RESULTS: Previously, we showed that COUP-TFI expression is enhanced in breast cancer compared to normal tissue. Here, we report that the CXCL12/CXCR4 signaling pathway, a crucial pathway in cell growth and migration, is an endogenous target of COUP-TFI in breast cancer cells. The overexpression of COUP-TFI in MCF-7 cells inhibits the expression of the chemokine CXCL12 and markedly enhances the expression of its receptor, CXCR4. Our results demonstrate that the modification of CXCL12/CXCR4 expression by COUP-TFI is mediated by the activation of epithelial growth factor (EGF) and the EGF receptor. Furthermore, we provide evidence that these effects of COUP-TFI increase the growth and motility of MCF-7 cells in response to CXCL12. Cell migration toward a CXCL12 gradient was inhibited by AMD3100, a specific antagonist of CXCR4, or in the presence of excess CXCL12 in the cell culture medium. The expression profiles of CXCR4, CXCR7, CXCL12, and COUP-TFI mRNA in 82 breast tumors and control non-tumor samples were measured using real-time PCR. CXCR4 expression was found to be significantly increased in the tumors and correlated with the tumor grade, whereas the expression of CXCL12 was significantly decreased in the tumors compared with the healthy samples. Significantly higher COUP-TFI mRNA expression was also detected in grade 1 tumors. CONCLUSIONS: Together, our mechanistic in vitro assays and in vivo results suggest that a reduction in chemokine CXCL12 expression, with an enhancement of CXCR4 expression, provoked by COUP-TFI, could be associated with an increase in the invasive potential of breast cancer cells.

Plasma-based analysis of ERBB2 mutational status by multiplex digital PCR in a large series of patients with metastatic breast cancer.

Erb-b2 receptor tyrosine kinase 2 (ERBB2)-activating mutations are therapeutically actionable alterations found in various cancers, including metastatic breast cancer (MBC). We developed multiplex digital PCR assays to detect and quantify ERBB2 mutations in circulating tumor DNA from liquid biopsies. We studied the plasma from 272 patients with hormone-receptor-positive, human epidermal growth factor receptor 2-negative (HR+/HER2-) MBC to detect 17 ERBB2 mutations using a screening assay. The assay was developed on the three-color Crystal dPCR™ naica® platform with a two-step strategy for precise mutation identification. We found that nine patients (3.3%) harbored at least one ERBB2 mutation. The mutation rate was higher in patients with lobular histology (5.9%) compared to invasive breast carcinoma of no special type (2.6%). A total of 12 mutations were found with the following frequencies: L755S (25.00%), V777L (25.00%), S310Y (16.67%), L869R (16.67%), S310F (8.33%), and D769H (8.33%). Matched tumor samples from six patients identified the same mutations with an 83% concordance rate. In summary, our highly sensitive multiplex digital PCR assays are well suited for plasma-based monitoring of ERBB2 mutational status in patients with MBC.

Development of sensitive and robust multiplex digital PCR assays for the detection of ESR1 mutations in the plasma of metastatic breast cancer patients.

BACKGROUND: Early detection of ESR1 mutations is a key element for better personalization of the management of patients with HR+/HER2- Metastatic Breast Cancer (MBC). Analysis of circulating tumor DNA from liquid biopsies is a particularly well-suited strategy for longitudinal monitoring of such patients. MATERIALS AND METHODS: Using the naica® three-color digital PCR platform, we developed a screening assay allowing the detection of 11 ESR1 mutations and designed a sequential strategy for precise mutation identification. We then applied this strategy in the analysis of plasma circulating cell-free DNA from 109 HR+/HER2- MBC patients and performed a double-blind comparison study on a subset of patients with the multiplex assay used at the Institut Curie (IC) for the PADA-1 study. RESULTS: Thirty-one patients (28.4%) harboured at least one ESR1 mutation, with the following frequencies: D538G (41.03%), Y537S (25.64%), E380Q (10.26%), Y537N (10.26%), "(536-540)" (7.69%), Y537C (2.56%), and L536R (2.56%). The presence of ESR1 mutation(s) was significantly associated with liver metastases (p = 0.0091). A very good agreement (91%) was observed with the IC assay. CONCLUSION: Our assays have proven to be robust and highly sensitive and are very well-suited for monitoring ESR1 mutations in the plasma of MBC patients.

Sentinel lymph node analysis in breast cancer: contribution of one-step nucleic acid amplification (OSNA).

One-step nucleic acid amplification (OSNA, Sysmex, Kobe, Japan) offers an excellent opportunity for accurate exhaustive sentinel lymph node (SLN) examination in breast cancer patients. Calibrated with conventional postoperative histology, this molecular technique yields comparable results intraoperatively, expressed as micrometastasis, macrometastasis or no metastasis depending on the CK19 mRNA copy number amplified in SLN lysates. We applied OSNA to detect metastasis in 810 SLNs from 367 patients with early stage breast cancer. We compared the rate of OSNA-positive SLNs in patients with invasive breast cancer (< 2 cm) versus the rate observed in a historical cohort using conventional histological examination of SLNs. No significant difference was observed, the OSNA assay was positive in 24.4% of patients, compared with positive histology in 24.8% in the historical cohort if including patients with isolated tumour cell (ITC) and in 23.4% excluding them. Opportunities for optimised patient management using OSNA are discussed: intraoperative detection of OSNA-positive SLNs enables axillary lymph node dissection (ALND) during the same procedure; standard OSNA techniques enable the establishment of homogeneous groups based on examination of whole SLNs for valid comparisons between different centres.

A cell-based system combined with flow cytometry to evaluate antibody responses against SARS-CoV-2 transmembrane proteins in patients with COVID-19.

This protocol describes a flow cytometry approach to evaluate antibody responses against SARS-CoV-2 transmembrane proteins in COVID-19-positive patient sera samples without the need of specific laboratory facilities for viral infection. We developed a human-cell-based system using spike-expressing HEK293T cells that mimics membrane insertion and N-glycosylation of viral integral membrane proteins in host cells. This assay represents a powerful tool to test antibody responses against SARS-CoV-2 variants and vaccine effectiveness. For complete details on the use and execution of this protocol, please refer to Martin et al. (2021).

SARS-CoV-2 integral membrane proteins shape the serological responses of patients with COVID-19.

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic has elicited a unique mobilization of the scientific community to develop efficient tools to understand and combat the infection. Like other coronavirae, SARS-CoV-2 hijacks host cell secretory machinery to produce viral proteins that compose the nascent virions; including spike (S), envelope (E), and membrane (M) proteins, the most exposed transmembrane proteins to the host immune system. As antibody response is part of the anti-viral immune arsenal, we investigate the immunogenic potential of S, E, and M using a human cell-based system to mimic membrane insertion and N-glycosylation. Both S and M elicit specific Ig production in patients with SARS-CoV-2. Patients with moderate and severe diseases exhibit elevated Ig responses. Finally, reduced Ig binding was observed with spike G614 compared to D614 variant. Altogether, our assay points toward an unexpected immune response against M and represents a powerful tool to test humoral responses against actively evolving SARS-CoV-2 variants and vaccine effectiveness.

Impact of EPclin on adjuvant therapeutic decision making and comparison of EPclin to the PREDICT tool.

PURPOSE: Genomic signatures, such as EndoPredict®, may help clinicians to decide which adjuvant treatment is the most appropriate. METHODS: We propose the EndoPredict® assay for unclear cases of adjuvant treatment in patients treated in our comprehensive cancer center. We prospectively and retrospectively report the decision of adjuvant treatment before and after the EndoPredict® assay, respectively, compared to the PREDICT's tool scores. RESULTS: From November 2016 to March 2019, 159 breast cancer tumors were analyzed and presented before and after the EndoPredict® assay. Before the EndoPredict® results, clinicians recommended chemotherapy for 57 patients (57/159, 36%). A total of 108 patients (108/159, 68%) were classified as EPclin high-risk score. There was only a slight agreement between clinicians' decisions and EPclin risk score. The EPclin score led to 37% changes in treatment (59/159); chemotherapy was favored in 80% of cases (47/59). The PREDICT tool recommended chemotherapy for 16 high-risk patients (16/159, 10%). CONCLUSION: Although genomic tests were developed in order to de-escalate adjuvant treatment, in our comprehensive cancer center the use of the EndoPredict® assay led to an increase in prescribed chemotherapy.

Development of multiplex digital PCR assays for the detection of PIK3CA mutations in the plasma of metastatic breast cancer patients.

With the approval of new therapies targeting the PI3K pathway, the detection of PIK3CA mutations has become a key factor in treatment management for HR+/HER2- metastatic breast cancer (MBC). We developed multiplex digital PCR (dPCR) assays to detect and quantify PIK3CA mutations. A first screening assay allows the detection of 21 mutations, with a drop-off system targeting the 542-546 hotspot mutations combined with the simultaneous detection of N345K, C420R, H1047L and H1047R mutations. In the case of a positive result, a sequential strategy based on other assays that we have developped allows for precise mutation identification. Clinical validity was determined by analyzing plasma circulating free DNA (cfDNA) from 213 HR+/HER2- MBC samples, as well as DNA extracted from 97 available matched tumors from 89 patients. Our assays have shown reliable specificity, accuracy and reproducibility, with limits of blank of three and four droplets for the screening assay. Sixty-eight patients (32%) had at least one PIK3CA mutation detectable in their plasma, and we obtained 83.1% agreement between the cfDNA analysis and the corresponding tumors. The high sensitivity and robustness of these new dPCR assays make them well-suited for rapid and cost-effective detection of PIK3CA mutations in the plasma of MBC patients.

An EMT-primary cilium-GLIS2 signaling axis regulates mammogenesis and claudin-low breast tumorigenesis.

The epithelial-mesenchymal transition (EMT) and primary ciliogenesis induce stem cell properties in basal mammary stem cells (MaSCs) to promote mammogenesis, but the underlying mechanisms remain incompletely understood. Here, we show that EMT transcription factors promote ciliogenesis upon entry into intermediate EMT states by activating ciliogenesis inducers, including FGFR1. The resulting primary cilia promote ubiquitination and inactivation of a transcriptional repressor, GLIS2, which localizes to the ciliary base. We show that GLIS2 inactivation promotes MaSC stemness, and GLIS2 is required for normal mammary gland development. Moreover, GLIS2 inactivation is required to induce the proliferative and tumorigenic capacities of the mammary tumor­initiating cells (MaTICs) of claudin-low breast cancers. Claudin-low breast tumors can be segregated from other breast tumor subtypes based on a GLIS2-dependent gene expression signature. Collectively, our findings establish molecular mechanisms by which EMT programs induce ciliogenesis to control MaSC and MaTIC stemness, mammary gland development, and claudin-low breast cancer formation.

One-step nucleic acid amplification for detecting lymph node metastasis of head and neck squamous cell carcinoma.

BACKGROUND: In head and neck squamous cell carcinoma (HNSCC) 30% of cN0 patients have occult metastasis. LN invasion is a major prognostic factor. Sentinel lymph node (SLN) is an option for cN0 neck management. One-step nucleic acid amplification (OSNA) used to analyze SLN in breast cancer is also a candidate to get more reliable intraoperative HNSCC lymph node (LN) staging. OBJECTIVE: To compare OSNA analysis to pathological analysis in cN0 HNSCC. MATERIALS AND METHODS: 157 LN from 26 cN0 HNSCC patients were prospectively analyzed (6.3LN/patient). Exclusion criteria were previous surgery or radiotherapy. Each node was cut into 4 equal pieces alternatively sent to pathological analysis and OSNA technique. IHC CK19 was performed on the primary tumor biopsy and RT-qPCR of CK19, PVA and EPCAM on the LN lysate of discordant cases. RESULTS: OSNA was able to provide intraoperative result in all patients. OSNA detected 21 metastases. There were 139 concordant LN (88.5%). There were 18 initial discordant LN (11.5%), 13 (8.3%) were OSNA positive/pathological analysis negative, 5 (3.2%) were OSNA negative/pathological analysis positive. After elimination of allocation bias, false negative rate was 1.3%, sensitivity and specificity were 90% and 95.6%, PPV and NPV were 75% and 98.5%. CONCLUSION: Our results suggest that OSNA should be considered to improve SNB analysis both for increasing micro metastasis diagnosis and offer extemporaneous results. Study registered under clinicaltrials.gov database number NCT02852343.

Tumor analysis: freeze-thawing cycle of triple-negative breast cancer cells alters tumor CD24/CD44 profiles and the percentage of tumor-infiltrating immune cells.

OBJECTIVE: The use of novel methods to characterize living tumor cells relies on well-conceived biobanks. Herein, we raised the question of whether the composition of fresh and freeze/thawed dissociated tumor samples is comparable in terms of quantitative and qualitative profiling. RESULTS: Breast cancer is a heterogeneous disease, encompassing luminal A and B, basal/triple-negative breast cancer (TNBC), and ERBB2-like tumors. We examined living cells dissociated from TNBC and found that a classical freeze/thaw protocol leads to a marked reduction in the number of CD45-CD44LowCD24Low tumor cells. This, in turn, changed the percentage of tumor cells with certain CD44/CD24 expression patterns and changed the percentage of tumor-infiltrating immune cells. These cryopreservation-driven alterations in cellular phenotype make it impossible to compare fresh and frozen samples from the same patient directly. Moreover, the freeze/thaw process changed the transcriptomic signatures of triple-negative cancer stem cells in such a manner that hierarchical clustering no longer ranked them according to expected inter-individual differences. Overall, this study suggests that all analyses of living tumor cells should be conducted only using freshly dissociated tumors if we are to generate a robust scoring system for prognostic/predictive markers.

Inhibition of IRE1 RNase activity modulates the tumor cell secretome and enhances response to chemotherapy.

Triple-negative breast cancer (TNBC) lacks targeted therapies and has a worse prognosis than other breast cancer subtypes, underscoring an urgent need for new therapeutic targets and strategies. IRE1 is an endoplasmic reticulum (ER) stress sensor, whose activation is predominantly linked to the resolution of ER stress and, in the case of severe stress, to cell death. Here we demonstrate that constitutive IRE1 RNase activity contributes to basal production of pro-tumorigenic factors IL-6, IL-8, CXCL1, GM-CSF, and TGFß2 in TNBC cells. We further show that the chemotherapeutic drug, paclitaxel, enhances IRE1 RNase activity and this contributes to paclitaxel-mediated expansion of tumor-initiating cells. In a xenograft mouse model of TNBC, inhibition of IRE1 RNase activity increases paclitaxel-mediated tumor suppression and delays tumor relapse post therapy. We therefore conclude that inclusion of IRE1 RNase inhibition in therapeutic strategies can enhance the effectiveness of current chemotherapeutics.

Elaboration of a nomogram to predict nonsentinel node status in breast cancer patients with positive sentinel node, intraoperatively assessed with one step nucleic amplification: Retrospective and validation phase.

BACKGROUND: Tumor-positive sentinel lymph node (SLN) biopsy results in a risk of non sentinel node metastases in micro- and macro-metastases ranging from 20 to 50%, respectively. Therefore, most patients underwent unnecessary axillary lymph node dissections. We have previously developed a mathematical model for predicting patient-specific risk of non sentinel node (NSN) metastases based on 2460 patients. The study reports the results of the validation phase where a total of 1945 patients were enrolled, aimed at identifying a tool that gives the possibility to the surgeon to choose intraoperatively whether to perform or not axillary lymph node dissection (ALND). METHODS: The following parameters were recorded: Clinical: hospital, age, medical record number; Bio pathological: Tumor (T) size stratified in quartiles, grading (G), histologic type, lymphatic/vascular invasion (LVI), ER-PR status, Ki 67, molecular classification (Luminal A, Luminal B, HER-2 Like, Triple negative); Sentinel and non-sentinel node related: Number of NSNs removed, number of positive NSNs, cytokeratin 19 (CK19) mRNA copy number of positive sentinel nodes stratified in quartiles. A total of 1945 patients were included in the database. All patient data were provided by the authors of this paper. RESULTS: The discrimination of the model quantified with the area under the receiver operating characteristics (ROC) curve (AUC), was 0.65 and 0.71 in the validation and retrospective phase, respectively. The calibration determines the distance between predicted outcome and actual outcome. The mean difference between predicted/observed was 2.3 and 6.3% in the retrospective and in the validation phase, respectively. The two values are quite similar and as a result we can conclude that the nomogram effectiveness was validated. Moreover, the ROC curve identified in the risk category of 31% of positive NSNs, the best compromise between false negative and positive rates i.e. when ALND is unnecessary (<31%) or recommended (>31%). CONCLUSIONS: The results of the study confirm that OSNA nomogram may help surgeons make an intraoperative decision on whether to perform ALND or not in case of positive sentinel nodes, and the patient to accept this decision based on a reliable estimation on the true percentage of NSN involvement. The use of this nomogram achieves two main gools: 1) the choice of the right treatment during the operation, 2) to avoid for the patient a second surgery procedure.

CD95L cell surface cleavage triggers a prometastatic signaling pathway in triple-negative breast cancer.

Triple-negative breast cancers (TNBC) lacking estrogen and progesterone receptors and HER2 amplification have a relatively high risk of metastatic dissemination, but the mechanistic basis for this risk is not understood. Here, we report that serum levels of CD95 ligand (CD95L) are higher in patients with TNBC than in other patients with breast cancer. Metalloprotease-mediated cleavage of CD95L expressed by endothelial cells surrounding tumors generates a gradient that promotes cell motility due to the formation of an unconventional CD95-containing receptosome called the motility-inducing signaling complex. The formation of this complex was instrumental for Nox3-driven reactive oxygen species generation. Mechanistic investigations revealed a Yes-Orai1-EGFR-PI3K pathway that triggered migration of TNBC cells exposed to CD95L. Our findings establish a prometastatic function for metalloprotease-cleaved CD95L in TNBCs, revisiting its role in carcinogenesis.