Bevacizumab in ovarian cancer: Clinical data and predictive and prognostic biomarkers.

The authors have nothing to report.

Worldwide, ovarian cancer (OC) ranks as the eighth most common cancer among women, accounting for an estimated 3.7% of all cancer cases and 4.7% of cancer‐related deaths in 2020. 1 Approximately 25% of OC diagnoses are hereditary: germline mutations in the BRCA1 and BRCA2 genes represent the most well‐established risk factors, 2 while a minor subgroup is attributable to other gene variants, including DNA mismatch repair genes (Lynch syndrome) and those involved in the DNA double‐strand break repair system, such as CHEK2, RAD51, BRIP1 and PALB2. 3 The main recognised environmental risk factors are nulliparity or infertility, obesity, hormonal replacement therapy and endometriosis, which is strongly associated with clear cell and low‐grade endometrioid and serous OC histotypes. 4 On the other hand, long‐term use of oral contraceptives (≥10 years) lowers risk by up to 60% 5 and four full‐term pregnancies are associated with a roughly 40% risk decrease. 6 The pathogenesis of OC is still unclear; however, it is well established that angiogenesis is a critical driver of OC progression, supporting tumour growth and extensively remodelling the tumour microenvironment (TME). Central to this process is the vascular endothelial growth factor (VEGF)–VEGFR signalling axis, where VEGF binding to VEGFR‐2 activates the PI3K/AKT and RAS/MAPK pathways, promoting endothelial cell proliferation, migration, survival and neovascularisation 7 (Figure  1 ). These angiogenic effects enhance tumour invasiveness by inducing epithelial‒mesenchymal transition (EMT), disrupting vascular integrity, and increasing permeability, factors that collectively contribute to ascites formation. 8 VEGF also induces VE‐cadherin phosphorylation, weakening endothelial junctions and further aggravating vascular leakage. 9 Molecular mechanisms of the vascular endothelial growth factor (VEGF)/VEGFR signalling axis and its intracellular pathways. Key ligands and receptors involved in angiogenesis, with a focus on pathways activated by the VEGF‐A/VEGFR‐2 axis and its inhibitors. Hypoxic conditions within the TME exacerbate angiogenesis through the stabilisation of hypoxia‐inducible factor‐1α (HIF‐1α), which upregulates VEGF and perpetuates a feed‐forward loop between hypoxia and neovascularisation. 10 , 11 HIF‐1α also plays a key role in immune evasion by upregulating programmed death cell‐ligand 1 (PD‐L1), suppressing T‐cell activity and promoting the recruitment of regulatory T cells and tumour‐associated macrophages. 12 Moreover, hypoxia‐induced glycolysis increases lactic acid production, acidifying the TME and impairing immune cell function. This acidic environment also activates matrix metalloproteinases, which degrade the vascular basement membrane and facilitate tumour invasion and metastasis. 13 Emerging evidence also suggests that homologous recombination deficiency (HRD) promotes genomic instability and tumour hypoxia, resulting in pro‐angiogenic signalling, while concurrent VEGF pathway inhibition can further exacerbate DNA damage and potentiate PARP inhibitor (PARPi) sensitivity. 14 Bevacizumab is a recombinant humanised IgG1 monoclonal antibody that specifically targets VEGF‐A isoforms, thereby inhibiting the VEGF‐mediated angiogenic signalling pathway. 15 This blockade leads to inhibition of new capillary formation by suppressing endothelial proliferation, migration and tube formation; reduction of VEGF‐A‐mediated autocrine survival signalling within tumour cells; induction of ‘vascular normalisation’, characterised by tightening of endothelial junctions, increased pericyte coverage, decreased vessel permeability, improved perfusion, alleviation of hypoxia and enhanced delivery and efficacy of concomitant therapies. 16 Nevertheless, there are some strategies that tumours develop to escape these therapies. Resistance to bevacizumab occurs through a spatiotemporal adaptation process where tumours bypass VEGF‐A blockade by activating redundant signalling pathways such as fibroblast growth factor (FGF), platelet‐derived growth factor (PDGF) and angiopoietins. To sustain growth, the TME recruits bone‐marrow‐derived cells and pericytes that stabilise alternative vascular networks. Furthermore, treatment‐induced hypoxia triggers aggressive mechanisms such as vessel co‐option by which tumour cells subvert pre‐existing healthy vasculature and the induction of EMT, which promotes invasion and metastasis. 17 The aim of this review is to describe the data available on the clinical use of the antiangiogenic drug bevacizumab and of the biomarkers that have been investigated as possible prognostic or predictive factors in patients receiving these therapies.

Great efforts by different research group have been dedicated to find predictive biomarkers of response in OC also using sample from large clinical trials (Table  3 and Figure  3 ). The target molecule was selected based on its biological relevance in OC angiogenesis, its mechanistic role in tumour growth and immune modulation, and the feasibility of its evaluation through translational and clinically applicable approaches. Below we summarise these results discussing the evidences on circulating, tissue and signature biomarkers and dedicate particular evidence to the work of the MITO group who recently designed and conducted the phase IV MITO16A/MaNGO‐OV2A clinical trial that aimed to specifically identify and/or validate prognostic biomarkers of bevacizumab‐treated OC patients. 61 Biomarkers evaluation in ovarian cancer. 292 171 Multiplex immunofluorescence Gene expression profiling Abbreviations: ADAM 17, a disintegrin and metalloproteinase 17; AGP, arabinogalactan‐proteins; Ang1, angiopoietin‐1; Ang2, angiopoietin‐2; bev: bevacizumab; Ca 125, cancer antigen 125; CD, cluster of differentiation; CXCL, C‒X‒C motif chemokine ligand; CXCR, C‒X‒C motif chemokine receptor; EGFR, epidermal growth factor receptor; ELISA, enzyme linked immunosorbent assay; FGF, fibroblast growth factor; FGFR, fibroblast growth factor receptor; FLT4, Fms‐related tyrosine kinase 4; HER2, human epidermal growth factor; HIF‐1, hypoxia‐inducible factor‐1; HRD, homologous recombination deficiency; IHC, immunohistochemistry; IL‐6, interleukin‐6; LC‒MS, liquid chromatography‒mass spectrometry; MIR, microRNA; MVD, microvessel density; NGS, next generation sequencing; NRP‐1, neuropilin‐1; OS, overall survival; PDGFR, platelet‐derived growth factor receptors; PD‐L1, programmed death‐ligand 1; PFS, progression‐free survival; qPCR, quantitative polymerase chain reaction; SMA, smooth muscle active; SNV, single nucleotide variant; TCGA, The Cancer Genome Atlas Program; Tie2, angiopoietin‐1 receptor; TP53, tumour protein 53; VEGF, vascular endothelial growth factor; YKL‐40, chitinase‐3‐like protein. Schematic overview of the ovarian cancer tumour microenvironment and biomarkers of bevacizumab response. Scheme representing the studied biomarkers (in circles) in the context of tumour angiogenesis in ovarian cancer. Green circles and red circles indicate biomarkers associated with good or poor response to bevacizumab, respectively. Mass spectrometry and immunoassays analyses on serum samples from 205 patients enrolled in the ICON7 trial allowed to identify three candidate biomarkers (the tyrosine kinase receptor FLT4 [VEGFR‐3], α1‐acid glycoprotein [AGP] and mesothelin), which were differentially expressed between responders and non‐responders. These biomarkers were not individually predictive of bevacizumab response, but gained a predictive value when combined with CA‐125 in a biomarker index. 62 However, further studies are required to validate the usefulness of this index. The levels of 15 angiogenesis‐associated proteins (Ang2, FGFb, HGF, PDGFbb, VEGFA, VEGFC, GCSF, IL8, KGF, PlGF, VEGFR1, VEGFR2, Ang1, Tie2 and VEGFD) were analysed in plasma samples from the same patients included in the ICON7 using multiplex enzyme linked immunosorbent assays (ELISAs). The Ang1/Tie2 axis is a key pathway that regulates vascular stability. A combination of Ang1 and Tie2 expression was identified as predictor of bevacizumab response, but with minimal benefit advantage. 63 Subsequent confirmatory analyses on 650 serum samples from 92 patients showed that Tie2 increase in combination with CA‐125 was shown to be a more effective biomarker. 64 Tie2 was also confirmed to be a valuable response biomarker for VEGF inhibitors in mCRC 65 and is currently being tested in a prospective non‐interventional biomarker study that will hopefully provide more definitive insights on the value of the Tie2 test to predict the response to bevacizumab. 66 This is consistent with the fact that VEGF inhibition often leads to a deregulation of the Tie2 axis. A single report that collected blood samples from primary epithelial ovaria cancer (EOC) patients at surgery showed that increased serum levels of VEGF‐C predicted higher response to bevacizumab, 67 but further studies are needed to confirm this hypothesis. Analysis of plasma levels of IL‐6, Ang‐2, OPN, SDF‐1 (CXCL12), VEGF‐D, IL‐6R and GP130, all vascular modulators, on samples from patients enrolled in GOG‐0218 trial, 18 revealed that high IL‐6 levels may define patients most likely to benefit from the addition of bevacizumab to standard chemotherapy. 68 This is in agreement with an important role of IL‐6 in defining the inflammatory TME in OC. 69 The proangiogenic secreted glycoprotein YKL‐40 is an emerging prognostic biomarker for different types of cancer and its plasma levels are increased in OC patients compared to healthy subjects, and even more increased in metastatic settings. 70 , 71 Low plasma levels of YKL‐40 predicted improved PFS and OS outcome in patients with advanced OC treated with bevacizumab, although these results need validation in bigger cohorts. 72 High NRP‐1 expression was proposed to predict poor prognosis in OC patients, consistent with the role of NRP‐1 as an enhancer of VEGF signalling. 73 As a tissue biomarker, it failed to predict bevacizumab response in a retrospective analysis of the GOG‐0218 clinical trial 74 ; however, research is ongoing to determine if soluble NRP‐1 could represent a circulating biomarker to predict bevacizumab response. 75 A tumour suppressor miRNA, miR‐200c, has been linked with EMT and resistance to chemotherapy. 76 Low level of miR‐200c in plasma can predict patients who can benefit from bevacizumab addition, but validation is necessary. 77 Predictive and prognostic biomarkers are under investigation also in tumour‐derived cfDNA from plasma and ascites of OC patients. 78 Ascites in particular has been found to contain abundant cfDNA and could represent a source of potential biomarkers. Tissue samples collected in the ICON7 trial identified VEGF‐A 165b as an ‘antiangiogenic’ isoform that could contribute to poor response to bevacizumab treatment. Patients with low levels of VEGF‐A 165b , detectable by IHC, showed improved OS when bevacizumab was added to standard platinum/paclitaxel‐based chemotherapy. 79 Using a tissue microarray prepared using a subset of samples from patients enrolled in ICON7 trial, it has been shown that high c‐MET/VEGFR‐2 co‐localisation on tumour tissue was shown to be associated with poorer survival outcomes in bevacizumab‐treated patients; moreover, from the same trial, patients bearing the specific VEGFR‐2 rs2305945 G/G variant had shorter PFS when treated with bevacizumab. 80 However, this result needs to be validated in larger cohorts. Consistently with the bevacizumab mechanism of action, tumoural VEGF‐A expression and microvessel density (MVD), evaluated by IHC using the expression of the endothelia marker CD31, showed potential predictive value in retrospective tumour biomarker analyses of the GOG‐0218 trial. 74 These data merit future confirmation from prospective randomised trial. Different other biomarkers were tested in samples collected from patients not included in randomised trials. Analyses of Ang‐2 expression on tumour tissue using Western blot and IHC showed that high expression correlates with significant benefit from bevacizumab therapy. 81 High levels of PDGFR‐beta and VEGFR‐2 have been correlated with platinum resistance and poor prognosis. 82 The same group has recently tried to associate stromal levels of these factors, analysed by tissue microarray IHC, with predictive value for bevacizumab treatment; consistent with the previous results, high expression of PDGFR‐beta was associated with shorter OS, but this was not specific to the group of bevacizumab‐treated patients. Bevacizumab treatment led to better prognosis in all patients, regardless of PDGFR‐beta levels. 83 VEGFR‐2 protein levels were almost absent in tumour samples, and this correlated with improved OS, but again not only in bevacizumab‐treated patients, suggesting that optimal bevacizumab response does not rely on VEGFR‐2 expression. 83 Vascular‐associated microRNAs, also known as angiomiRs, have been shown to regulate angiogenesis and metastasis. miR‐378 has been shown to promote angiogenesis in various cancers. In vitro experiments correlated with data from The Cancer Genome Atlas Program (TCGA) showed that low miR‐378 expression was predictive of response to combination bevacizumab and chemotherapy. 84 miR‐6086 has been shown to inhibit angiogenesis in OC by impacting the OC2/VEGFA/EGFL6 axis and could be used as a biomarker. 85 In addition to miRNAs, regulation of angiogenesis‐related processes by other epigenetic mechanisms has been gaining attention. Inhibition of the epigenetic modifiers BRD2/3/4 belonging to the BET family of proteins has been shown to regulate the response to antiangiogenic therapy. 86 Further investigation could be useful to identify biomarkers of response and resistance to bevacizumab. Indeed, a study reports that BET inhibitors demonstrated efficacy in overcoming resistance to radiotherapy in OC, suggesting the role of the BET‐regulated gene GNL3 as a biomarker for selecting personalised therapy. 87 Molecular signatures derived from gene expression profile (GEP) studies divided HGSOC in four different subtypes association with prognostic significance. 88 , 89 Application of these signatures to a subset of patients from the ICON7 trial found a correlation between molecular subtype 88 and response to anti‐angiogenic therapy, showing that the mesenchymal and proliferative subtype, those with poorest survival, have a better response to anti‐angiogenic therapy. 79 A study tried to correlate patient outcome with the expression of the fibroblast growth factor receptors and their ligands (FGFRs/FGFs), a family of proteins involved in angiogenic signalling. A signature based on the expressions of FGFR1, FGFR4 and FGF19 was able to predict favourable prognosis for bevacizumab‐treated patients. 90 Finally, genetic alterations of EGFR or HER2 were found to be associated with poor response to bevacizumab plus chemotherapy. 91 Overall, our literature search did not identify validated biomarkers of response to bevacizumab, which could be ready for clinical application. Serum biomarkers would represent the most promising, allowing a minimally invasive detection from patients’ plasma. However, all the studies we examined suffer from the lack of external validation, either in bigger cohorts or in biomarker‐driven prospective trials. Moreover, lack of consistency could be also due to poor reproducibility: the assays used in single laboratories are not standardised in terms of cut‐offs or statistical analyses.

The most recent literature reviewed shows two main limits for biomarkers validation: rarely the same biomarker was tested using the same approach in different studies by independent researchers and only few studies were based on analyses of samples from patients included in clinical trials with translational endpoints. The adding value to have patients with similar clinical characteristics and homogenously treated in a controlled clinical trial should not be underestimated. The MITO group tried to deal with these two limitations by designing and conducting a prospective phase IV trial whose principal aim was to validate biomarkers emerged from smaller cohorts of patients or based on biological rationales from preclinical studies. Many of the initially promising biomarkers did not retain statistical significance once the analyses were adjusted for multiple testing using appropriate shrinkage methods. A rigorous statistical correction is therefore essential to ensure that only truly robust biomarkers progress to further validation, ultimately preserving time, biological samples and financial resources and to avoid overestimating the relevance of individual signals and generate false‐positive findings. The identification of robust prognostic and predictive biomarkers could enable a more targeted and effective use of bevacizumab, enhancing the therapeutic potential of its latest combinations with innovative agents, including ADCs, ICIs and PARPi. Biomarkers such as HRD, for example, may help guide treatment selection and sequencing, optimising the use of targeted therapies and maximising the clinical benefit of these emerging combinatorial strategies. In conclusion, we want to point out that the use of novel technologies such as spatial proteomics and transcriptomics along with the more and more widespread use of nucleic sequences produce compelling and potentially relevant data and could be of full clinical utility when integrated with novel approaches of ML and AI. In this context, spatial transcriptomics and multiplexed proteomics could enable the spatially resolved identification of angiogenic, stromal and immune signatures, which, when integrated into AI‐driven models, may significantly improve biomarker discovery and bevacizumab response prediction in future trials. 110 This could allow also the integration of multiple biomarkers to generate ‘complex signatures’. However, these new technologies are still at research grade, and accurate standardisation would be required for the application of these techniques in diagnostic settings. Based on these considerations, we believe that the future of cancer research requires the complete integration of multidisciplinary expertise that is particularly possible in the frame of large cooperative clinical and translational research groups.

To establish a standardised system to identify biomarkers of bevacizumab response, the MITO group in Italy has specifically conceived the MITO16A/MaNGO OV‐2 trial. The multicentre, phase IV, single‐arm MITO16A/MaNGO OV‐2 trial was designed to explore molecular and clinical biological factors allowing to identify OC patients who could benefit from the addition of bevacizumab to standard chemotherapy in first line. 61 The biomarker‐driven trial enrolled 398 patients who all received bevacizumab with first‐line chemotherapy. One limitation of the study was the absence of a chemotherapy‐only control arm, which allowed for the identification of prognostic biomarkers within a bevacizumab‐treated cohort, but prevented the evaluation of predictive biomarkers. Tissue samples collection was centralised at the INT G. Pascale of Naples that supervised the quality controls and performed tissue processing to build tissue micro‐array (TMA), to extract nucleic acid and provide final tissue slides to investigators as necessary (Figure 4 ). Biomarkers analyses included the evaluation of protein expression on TMA or tissue slides, depending on the biomarker in study, the evaluation of gene and microRNA expression and the analyses of DNA somatic alterations including HRD. To ensure the reliability of the data, the MITO‐group researchers agreed to perform the wet analyses locally and have a central independent statistical evolution of the results that followed rigorous statistic pipelines and avoided data torturing (Figure  4 ). Schematic workflow of the MITO16A/Mango OV2 trial. The MITO clinical and translational network jointly designed the translational endpoints of the trial looking at the available information and personal unpublished data. After the completion of patients’ enrollment, a review of the anticipated biomarkers has been done. Samples collection and processing (tissue micro‐array and whole slide preparation, nucleic acid extraction) have been centralised to ensure quality control of used material. The clinical database was also centralised. Processed tissues were distributed to the peripheral centre that performed locally the analyses of selected biomarkers. All clinical and molecular data were then processed for prognostic and statistical analyses by a centralised group of statisticians in blind to ensure the reliability of the results. The first attempt was to confirm a possible prognostic role for angiogenesis‐related molecules such as MVD evaluated using CD31 expression, α‐SMA, VEGFA, VEGFR‐2 and HIF‐1α that were assessed on tissue slides. However, once correlated with clinical variables and patients’ outcome, these markers lost significance. 92 Due to the design of the study in which all patients received bevacizumab treatment we could not define the predictive value of tissue VEGFA and MVD proposed by Bais et al. 74 Analysis of an angiogenesis‐related microRNA signature, and in particular miR‐484, a regulator of VEGFB and VEGFR‐2 pathways, 93 did not show significant correlation with MVD, although high levels of miR‐484 were observed in VEGF‐B‐negative samples, confirming previous observations. 93 As a possible hint for future investigation it has been observed that the association of high miR‐484 levels and high VEGF‐B expression was associated with longer PFS. 92 Regarding the expression of VEGFR‐2, the work confirmed that this receptor is generally not expressed by tumour cells, as observed by others, 83 and that therefore the use of TMAs is not a valuable tool to evaluate its expression in OC. 92 A disintegrin and metalloprotease 17 (ADAM17) is involved in OC progression and cisplatin resistance. 94 A low ADAM17 score in the examined patients was associated with better survival; however, the prognostic value lost significance after statistical correction for multiple testing and could be prognostic only in the subgroup of patients without residual disease at baseline. 95 ADAM17 mediates activation of EGFR signalling through proteolytic cleavage. EGFR overexpression in OC has been associated with poor prognosis. Combination of GEP and IHC studies demonstrated that strong and homogeneous EGFR membrane staining by IHC were associated with the worst outcome after bevacizumab treatment. This group of patients was correlated with activation of EGFR pathways and lack of angiogenic‐related molecular features. 96 Next‐generation sequencing (NGS) sequencing and IHC studies were used to evaluate the possible prognostic role of TP53 , the gene most frequently mutated in OC. TP53  mutational status was assessed by NGS and correlated with p53 expression evaluated by IHC. In the MITO16a case material, the presence of  TP53  mutations (in particular unclassified missense mutations) was associated with increased OS benefit in patients receiving bevacizumab in addition to standard chemotherapy independently of clinicopathological characteristics. On the other side, TP53 status evaluated by IHC did not have prognostic significance in the same case group. 97 These results highlight the possible different significance of diverse TP53 mutation and warrant future confirmatory investigations. NGS approaches were also used to evaluate the prognostic significance of HRD in a subset of patients enrolled in the MITO16A trial using commercial and academic approaches. All collected data demonstrated that evaluation of HRD using NGS approaches had prognostic significance also in patients treated with bevacizumab in first line settings. 98 , 99 , 100 On the other side, the evaluation of HRD using a functional approach based on the evaluation of RAD51 expression by immunofluorescence 101 did not provide conclusive prognostic indication in the studied population. However, the combined evaluation of functional and genomic HRD coupled with the expression of miR‐506 may be used to further refine HRD status and identify a group of patients with particular good prognosis. 100 A growing body of evidence supports the idea that bevacizumab can modulate the complex tumour immune microenvironment in OC. Therefore, immune‐related molecules could be relevant biomarkers of the response to bevacizumab treatment. 102 The CXCR4‒CXCL12 chemokine axis plays a significant role in OC by promoting cell proliferation, migration, invasion and metastasis. It also contributes to reshape the TME, mainly by altering the immune responses. 103  Epithelial and stromal expression of CXCL12 and its receptors CXCR4 and CXCR7 was evaluated by IHC on 308 EOC samples included in the MITO16A TMA. Among these proteins, only high epithelial CXCL12 was negatively associated with PFS and OS, although the statistical significance was lost after adjusting for overfitting. 104 Full sections from 292 patients were analysed by a multiplex immunofluorescence (MIF) approach applied to the MITO16A case material. This allowed the evaluation of the levels and the spatial localisation of programmed death‐ligand 1 (PD‐L1), tumour cells (cytokeratins), T cells (CD8) and monocyte/macrophage populations on a single tissue slide. The absence of PD‐L1‐positive cells in the tumour and the presence of PD‐L1‐positive cells in the stroma were associated with a better prognosis for patients treated with chemotherapy plus bevacizumab, although the associations lost significance after statistical correction. 105 The prognostic value of the immune infiltrate was further examined coupling spatial MIF analysis of CD8 + and CD68 + cells and molecular stratification of patients through gene expression profiling. The results suggest that immune‐infiltrated tumours responded less to bevacizumab treatment, although this finding needs to be confirmed with prospective trials. 106 If validated in future prospective investigations, this observation would hold substantial translational significance. Tumour immune infiltration, particularly by CD8 + T lymphocytes, is a well‐established positive prognostic marker in OC patients treated with bevacizumab‐free regimens. Therefore, systematic assessment of the immune infiltrate may enable oncologists to stratify patients according to the likelihood of deriving clinical benefit from the addition of bevacizumab, while also identifying those for whom such treatment may be ineffective or even detrimental with respect to PFS. These works demonstrated the feasibility and utility of using MIF in large clinical trials in OC and support the possibility that spatial distribution of single biomarkers could have different prognostic value in cancer, adding complexities to the field. For these reasons we believe that machine learning (ML) and artificial intelligence (AI) approaches might help in the evaluation and integration of multiple variables providing a clearer and more reproducible picture of biomarkers prognostic/predictive roles in human cancer. In a recent multicentre study, deep‐learning models applied to digital histopathology slides accurately estimated a molecular signature associated with sensitivity to atezolizumab–bevacizumab, identifying patients with significantly improved PFS. This fast and scalable approach, potentially extensible to other solid tumours, including OC, suggests that AI‐driven models may support the development of non‐invasive predictive biomarkers for anti‐VEGF treatments such as bevacizumab. 107 We started to apply a ML approach on germline DNA variants associated with drug‐induced toxicities in 171 patients with OC enrolled in the MITO16A trial and demonstrated that ML identified a polygenic toxicity risk score able to better categorise the patients into high‐ and low risk of drug toxicities. 108 Several other studies have already shown concrete applications of ML/AI in biomarker prediction and clinical decision support, such as radiomic models capable of predicting BRCA/HRD and neural networks that classify microenvironmental immune patterns with prognostic value. This evidence further supports the utility of advanced computational approaches in integrating heterogeneous variables and improving the accuracy of patient stratification. 109 In conclusion, despite the limitations of the study, the analyses of biomarkers from the MITO16A/MaNGO OV‐2 trial allowed to identify key factors affected by bevacizumab treatment. The collected data can represent a valuable resource for the scientific community, supporting the work of other researchers to focus on specific pathways and identify predictive biomarkers of bevacizumab response in OC patients. In addition, the robust statistical pipeline exploited by the MITO group represents an example of rigorous statistical analysis of clinical biomarkers.

In patients with platinum‐resistant relapse, defined as disease recurrence occurring within 6 months after the last cycle of PBCT received, the AURELIA trial evaluated the addition of bevacizumab to chemotherapy, either liposomal doxorubicin, weekly paclitaxel or topotecan, compared with chemotherapy alone in patients with advanced platinum‐resistant OC. This study has shown a significant improvement, with a mPFS of 6.7 months in patients receiving the combination compared to 3.4 months in the arm of chemotherapy alone (HR: .48; p  < .001), with a manageable safety profile, although predictable adverse events. 31 Taken together, the studies discussed highlight the pivotal role of bevacizumab in the treatment of OC across different clinical settings. A true advancement in the field would be the identification of predictive biomarkers, which could enable a more precise selection of patients most likely to benefit from bevacizumab therapy. The combination of bevacizumab with new generation drugs is a topic of increasing interest. Recent evidence leads to the hypothesis that bevacizumab in combination with antibody‒drug conjugates (ADCs) could improve therapeutic effectiveness in OC using a complementary mechanism of action. Anti‐VEGF drugs temporarily normalise tumour vasculature, lowering interstitial pressure and facilitating ADCs intratumoural delivery, potentially increasing payload penetration and target engagement 32 (Figure  2a ). Several ongoing trials are investigating the role of combination with bevacizumab both in heavily treated recurrent OC, mainly with mirvetuximab soravtansine 33 and luveltamab tazevibulin, 34 and in the frontline maintenance setting, in particular with mirvetuximab soravtansine 29 and trastuzumab deruxtecan. 35 Synergistic mechanism of bevacizumab and antibody‒drug conjugates (ADCs) and pembrolizumab in ovarian cancer (OC). (a) This diagram illustrates how bevacizumab promotes vascular normalisation, enhancing the intratumoural delivery and penetration of ADCs. (b) This figure illustrates how bevacizumab mediates vascular normalisation and promotes T‐cell infiltration into the tumour microenvironment. This mechanism amplifies the activity of immune checkpoint inhibitors such as pembrolizumab, amplifying the immunosuppressive microenvironment of OC. In addition, also the combination of bevacizumab with ICIs in OC is supported by a strong biological evidence: angiogenesis contributes to an immunosuppressive microenvironment, consequently VEGF blockade promotes vascular normalisation, enhances T‐cell infiltration and amplifies ICI activity 36 (Figure  2b ). The recent phase III trial KEYNOTE‐B96 showed improved PFS and OS with pembrolizumab plus chemotherapy ± bevacizumab in platinum‐resistant disease 37 and encourage further exploration of these combination strategies. Even in this setting, the identification of biomarkers could ensure more accurate patient selection. In the last 20 years, bevacizumab has been approved by FDA and EMA for the treatment of metastatic colorectal cancer (mCRC), non‐small cell lung cancer (NSCLC), renal cell cancer, breast cancer (BC) and OC. Many studies have tried to identify predictive biomarkers of response, but to date no one has been validated. Since the addition of bevacizumab to standard therapy has shown benefit, general population is still being treated without pre‐selection, but a predictive biomarker would avoid unnecessary costs and toxicities of treating patients who do not show upfront response. The biomarkers investigated in the context of bevacizumab therapy have been primarily selected based on their biological involvement in VEGF‐driven angiogenesis, vascular stabilisation, hypoxia signalling and modulation of the tumour immune microenvironment. These molecules are hypothesised to influence sensitivity or resistance to VEGF blockade by reflecting either the dependency of the tumour on angiogenic pathways or the activation of compensatory mechanisms that may limit the clinical efficacy of bevacizumab. Levels of VEGF‐A, the main target of bevacizumab, have been one of the first investigated biomarkers. Increased concentration of VEGF‐A in patient serum was generally correlated with poor prognosis in hepatocellular carcinoma, 38 OC, 39 BC, 40 colorectal cancer (CRC) 41 and gastric cancer (GC). 42 However, no study has been able to demonstrate a predictive value for bevacizumab therapy response or to validate this finding in large cohorts in prospective trials, 43 , 44 , 45 , 46 , 47 concluding that VEGF‐A levels can be useful as a prognostic rather than a predictive biomarker. VEGF‐A alternative splicing generates several isoforms with different biological roles in angiogenesis; some studies have suggested that the analysis of specific isoforms or the ratio among them and total VEGF‐A levels could identify more specific biomarkers of bevacizumab efficacy in mCRC 48 , 49 and NSCLC. 50 Moreover, VEGF gene polymorphisms have been investigated in CRC but no predictive impact was demonstrated. 51 , 52 Other angiogenic factors were also investigated, but no clear biomarker was identified. BC patients with low tumour levels of the angiogenesis regulators delta‐like canonical notch ligand 4 (Dll4), VEGF‐C and neuropilin‐1 (NRP‐1) may benefit from bevacizumab addition to chemotherapy, 53 while tumour NRP‐1 levels were found to be good biomarker candidates in advanced GC. 45 High tumour expression of the sprouting enhancer APLN 54 and high plasma levels of placental growth factor (PlGF) and VEGF‐D 55 were predictive of failure to respond to bevacizumab treatment in mCRC. A signature based on plasma levels of angiopoietin like 4 (ANGPTL4), hepatocyte growth factor (HGF) and VEGF‐A 121 isoform could predict benefit from bevacizumab addition in mCRC. 56 Another study identified a plasma signature consisting of high levels of epidermal growth factor and macrophage‐derived chemokine and low levels of IL‐10, IL‐6 and IL‐8 as associated with better response to bevacizumab in mCRC. 57 On the other hand, angiogenic plasma biomarkers such as bFGF, E‐selectin, ICAM‐1, PlGF, VEGFR‐1 and VEGFR‐2 were not predictive of bevacizumab response in NSCLC. 46 Other factors from the TME were investigated, but the studies were only able to demonstrate associations and not biomarkers with predictive value. A signature with high expression of the tumor growth factor‐β (TGF‐β) superfamily member activin A, IL1β, and the urokinase plasminogen activator surface receptor correlated with response to bevacizumab treatment as monotherapy in metastatic melanoma. 58 hERG1 potassium channels regulate the secretion of VEGF‐A and the process of neoangiogenesis; the presence of hERG1 in the tumour, together with the presence of the active form of HIF‐2α, was associated with lower risk of progression when bevacizumab was added to chemotherapy. 59 VEGF stimulates the production of nitric oxide (NO). Monitoring of NO serum levels in 15 NSCLC patients before and during bevacizumab treatment demonstrated that reduced NO under bevacizumab was associated with a longer patients’ PFS, suggesting that NO can represent a dynamic biomarker of the response to bevacizumab. Of course larger series of patients and confirmatory studies are needed. 60 Taken together, these reports indicate that there is no consensus on the role of angiogenic‐ and microenvironment‐related molecules in predicting the response to bevacizumab. The major limitations for most of the studies consist in the small sample size and lack of independent validation in subsequent work.

S.P. received honoraria from Roche, AZ, MSD, GSK and research funding from Roche, AZ, MSD, GSK and Pfizer. M.R.L., E.P., G.B., C.D.A., D.L. and F.B.V. declare they have no conflicts of interest.