cancers split 2

Med-MDPI/cancers_2/cancers-14-05-01129.txt

@@ -0,0 +1 @@
+In parallel to the successful clinical implementation of PARP1/2 inhibitors as anti-cancer drugs, which interfere with the DNA repair machinery, these small molecule agents have also gained attention as vehicles for molecular imaging and radiotherapy. In this review article, we summarize the development and preclinical evaluation of radioactively-labelled PARP inhibitors for positron emission tomography (PET) for many applications, such as selecting patients for PARP inhibitor treatment, response prediction or monitoring, and diagnosis of tumors. We report on early clinical studies that show safety and feasibility of PARP-imaging in humans. In addition, we summarize the latest developments in the field of PARP-targeted radiotherapy, where PARP inhibitors are studied as vehicles to deposit highly cytotoxic radioisotopes in close proximity to the DNA of tumor cells. Lastly, we look at synthetic strategies for PARP-targeted imaging and therapy agents that are compatible with large scale production and clinical translation. Since it was discovered that many tumor types are vulnerable to inhibition of the DNA repair machinery, research towards efficient and selective inhibitors has accelerated. Amongst other enzymes, poly(ADP-ribose)-polymerase 1 (PARP1) was identified as a key player in this process, which resulted in the development of selective PARP inhibitors (PARPi) as anti-cancer drugs. Most small molecule PARPi’s exhibit high affinity for both PARP1 and PARP2. PARPi are under clinical investigation for mono- and combination therapy in several cancer types and five PARPi are now clinically approved. In parallel, radiolabeled PARPi have emerged for non-invasive imaging of PARP1 expression. PARP imaging agents have been suggested as companion diagnostics, patient selection, and treatment monitoring tools to improve the outcome of PARPi therapy, but also as stand-alone diagnostics. We give a comprehensive overview over the preclinical development of PARP imaging agents, which are mostly based on the PARPi olaparib, rucaparib, and recently also talazoparib. We also report on the current status of clinical translation, which involves a growing number of early phase trials. Additionally, this work provides an insight into promising approaches of PARP-targeted radiotherapy based on Auger and α-emitting isotopes. Furthermore, the review covers synthetic strategies for PARP-targeted imaging and therapy agents that are compatible with large scale production and clinical translation.Due to continuous exposure to DNA-damaging events and the resulting DNA lesions, such as single-strand breaks (SSBs) and double-strand breaks (DSBs), cells maintain their genomic stability through the activation of numerous DNA damage response mechanisms, such as Base Excision Repair (BER), Homologous Recombination (HR), classical and alternative Non-Homologous End Joining (NHEJ), Nucleotide Excision Repair (NER), and Mismatch Repair (MMR) [1,2,3]. In those DNA damage repair pathways, some of the 17-member family of poly(ADP-ribose)polymerase (PARP) proteins play an essential role. PARP1 (in the following referred to as PARP for simplicity), a multifunctional enzyme of 113 kDa, is the best known and most abundantly expressed family member with multiple functions in DNA repair, genomic stability, and cell death [4]. Different factors, such as high rates of genomic instability, mutational burden, and defects in other DNA repair pathways, such as homologous recombination (HR) (e.g., BRCA1/2 mutations), lead to frequent dependence of tumors on PARP1-mediated DNA repair and high PARP1 expression levels [5]. Consequently, inhibition of PARP-mediated DNA repair was discovered to be an efficient approach to selectively kill tumor cells, which resulted in the development of small molecule PARP inhibitors (PARPi) that bind to the NAD+ binding pocket of the catalytic domain of PARP1 and prevent poly(ADP)-ribosylation [6,7,8,9,10]. Most PARPi also bind to the lesser expressed close homologue PARP2 with high affinity, therefore they are often named PARP1/2 inhibitors, which we simplify as PARPi. Intense research efforts have produced a high number of highly affine and selective PARPi, which have advanced to clinical evaluation [11]. In addition to monotherapy, mostly involving patients with germline or somatic BRCA1/2 or other HR-associated mutations, combination therapies with chemo-, radio- or immunotherapy have been or are currently investigated in clinical studies [11]. Since 2014, four PARPi (olaparib, rucaparib, talazoparib, and veliparip) have been clinically approved by the Food and Drugs Administration (FDA) and/or the European Medicines Agency (EMA) and one PARPi is approved in China (pamiparib). Olaparib received its first approval in 2014 and is now approved as monotherapy for the treatment of certain types of breast, ovarian, fallopian tube, peritoneal, pancreatic, and prostate cancer, as well as in combination with bevacizumab for ovarian cancer [12,13]. Rucaparib has received approval for the treatment of certain prostate and ovarian cancer patients that have failed other therapies [14]. In 2019, niraparib was approved for treatment of recurrent epithelial ovarian, fallopian tube, or primary peritoneal cancers [15], which was recently followed by approval as first line therapy for the same cancers. Talazoparib is currently approved for patients with locally advanced and metastatic breast cancer with germline BRCA1/2 mutations [16]. The high number of ongoing studies will likely result in additional approvals in the future. It is now known that the anti-cancer activity of PARPi is mediated by several mechanisms, but to fully understand the resulting in vitro and in vivo efficacy, as well as resistance to PARPi, is an ongoing field of investigation. Initially, catalytic inhibition of the PARP-mediated DNA repair was considered the main mechanism for PARPi-induced cytotoxicity, leading to synthetic lethality in HR deficient tumors. However, catalytic inhibition alone could not fully explain the effects of PARPi therapy, since, e.g., PARPi with similar PARP1/2 affinity exhibits different potency and sensitivity to PARPi and does not always depend on HR status. PARP trapping was identified as an additional mechanism of PARPi, where the dissociation of PARP enzymes from chromatin is prevented by PARPi binding, leading to replication fork stalling and eventually collapse, if not resolved by other DNA repair mechanisms [17]. Subsequently, pronounced differences in trapping potential of known PARPi were discovered [18], which is discussed to explain their differences in in vivo therapeutic efficacy [19]. New studies also add to a better understanding of response biomarkers and resistance mechanisms to PARPi therapy [20,21]. In addition, histone parylation factor 1 (HPF1) was recently identified as an important co-factor in the induction of PARP1/2 mediated DNA repair [22]. HPF1, which forms a joint active site with PARP1 or PARP2, actually also modulates the binding affinity of some PARPi to PARP1, indicating that HPF1 might directly affect PARP inhibition and trapping and, therefore, response to PARPi [23]. Hence, these additional levels of complexity require improved strategies for patient selection for PARPi therapy. While the importance of PARPi in cancer therapy is steadily growing, identification of responders and non-responders is still challenging. Next to the requirement of PARP1 expression for PARPi sensitivity, a number of resistance mechanisms are known, which can, e.g., circumvent dependence on PARP-mediated DNA repair or directly affect PARPi binding [20,21,24]. Therefore, non-invasive determination of PARP expression and indirect or direct measurement of PARPi binding could be a promising approach for improved patient selection. Radiolabeled probes offer excellent opportunities to measure PARP expression directly and noninvasively in patients via positron emission tomography (PET) or single photon emission tomography (SPECT). PARP imaging agents could be used as companion diagnostics for PARPi therapy, i.e., to select patients and for non-invasive whole-body PARP imaging to predict and/or monitor the response to PARPi therapy. Furthermore, the widespread overexpression of PARP could be leveraged for diagnostic imaging of tumors that are otherwise difficult to image with standard radiotracers, such as [18F]FDG, e.g., brain cancer [25] and head and neck cancer [26]. In addition, it was also suggested that PARPi could act as intranuclear delivery vehicles for therapeutic radioisotopes, including α- and Auger-emitters.To explore these clinically relevant applications, a considerable number of radiolabeled PARPi have emerged, for the most part, in the last ten years. The majority of radiolabeled PARPi are based on the structures of olaparib (Figure 1) and rucaparib (Figure 2). Talazoparib was also recently radiolabeled, as well as a few other PARP-targeting molecules (Figure 3). While several comprehensive reviews have previously described preclinical development and translational efforts [25,27,28,29,30], the extensive publication of novel radiolabeled PARPi, as well as clinical study results of the translated PARPi ([18F]FTT and [18F]PARPi), just in the last three years, warrants a systematic overview of the current status in the rapidly expanding field of PARP imaging and therapy. We included all existing probes in this analysis to gain an understanding of the effects of the different structural modifications compared to the parent PARPi on affinity, selectivity, cell permeability, and pharmacokinetic properties. With the increasing number of clinical studies in mind, we also take a look at the challenges and opportunities regarding synthesis of radiolabeled PARPi in the required quantity and quality for clinical translation. The first radiolabeled olaparib analogue emerged in 2011, several years before olaparib’s first clinical approval for the treatment of BRCA-mutated ovarian cancer in 2014 [13]. Several strategies have been explored to attach radiolabels, including 18F and iodine isotopes, to the olaparib scaffold, yielding a number of different olaparib-based radiotracers, of whom one, [18F]PARPi, has reached clinical evaluation to date. The first radiolabeled olaparib derivative was synthesized by Weissleder and colleagues via an inverse-electron Diels–Alder cycloaddition. Accordingly, an 18F-labelled trans-cyclooctene (TCO) with a tetrazine-modified olaparib derivative reacted with [18F]BO, also called [18F]AZD2281 [31]. [18F]BO (IC50 = 17.9 ± 1.1. nM) was the first olaparib-based radiotracer to show successful in vivo imaging of breast and ovarian cancer xenografts and its uptake correlated with PARP1 expression [32,33]. However, no further studies were conducted with [18F]BO. Subsequently, a bimodal PARP imaging agent, carrying a radiofluorinated fluorescent dye was achieved via Lewis acid-assisted fluoride exchange with triflic anhydride [34]. Although [18F]PARPi-FL was successfully synthesized and utilized for in vivo imaging of glioblastoma xenografts, it was not further developed due to its rapid metabolic defluorination in vivo [35]. The chemical modifications to achieve [18F]BO and [18F]PARPi-FL led to a relatively large footprint and bulky structure compared to olaparib, very likely introducing significant changes for target binding and pharmacokinetic properties, but also to the cell penetration ability, which is essential to reach the intranuclear target. Therefore, it is important to note that subsequent radiolabeled olaparib derivatives were structurally more similar and closer in molecular weight to their parent PARPi. Replacing the cyclopropyl moiety of olaparib with a fluorobenzene ring led to the radiochemically stable [18F]PARPi, which was developed in the Reiner lab in 2015. A large body of preclinical work subsequently explored the utility of [18F]PARPi for a variety of potential applications, such as diagnosis of brain, head, and neck cancers [26,36,37], quantification of PARPi target engagement [38], efficacy assessment of PARPi treatment, and for differentiation between malignant and non-malignant lesions in lymphomas and gliomas [39,40]. This tracer has also been clinically translated, which is detailed in Section 4. In vitro experiments showed a similar affinity and selectivity profile of [18,19F]PARPi and olaparib (IC50 value of [19F]PARPi: 2.83 nM, olaparib: 5 nM) [37,38], supporting that loss of the cyclopropyl moiety did not affect these properties, aligning with previous reports stating that it mainly increased oral bioavailability [7,41]. Recently, Wilson et al. suggested a simplified and faster two-step, one-pot radiosynthesis with a radiochemical yield of up to 9.6%, compared to the originally published multistep, multi-pot procedure to potentially facilitate production for clinical studies [42].[18F]PARPi was initially evaluated for glioblastoma imaging. In subcutaneous and orthotopic U251 MG xenograft models, the authors found a tumor uptake of 1.8–2.2% ID/g with high specificity (>85% blockable), but only a very low brain uptake, resulting in a tumor-to-brain ratio of 55, providing high signal to noise contrast [37]. [18F]PARPi was also able to clearly delineate gliomas with PET imaging in a genetically engineered mouse model, with higher accuracy than [11C]Choline and [18F]Fluorothymidine. The uptake correlated with PARP1 expression and was validated to be tumor specific in blocking experiments, autoradiography, and by using a fluorescent analogue (PARPi-FL) for microscopic evaluation [25]. The ability of [18F]PARPi to penetrate into areas of the brain inaccessible to high molecular weight FITC-Dextran in the genetically engineered glioma mouse model suggests blood-brain barrier (BBB) penetration by the tracer [43], which could enable tumor imaging even if the BBB is not compromised by the tumor. Head and neck cancer imaging was identified as another potential application for [18F]PARPi. These investigations followed studies showing PARP1 overexpression in oral and oropharyngeal cancer [44,45]. The group previously showed feasibility of diagnostic and intraoperative oral cancer imaging using the fluorescent PARP imaging agent PARPi-FL preclinically and clinically [44,45,46,47]. In a recent study, [18F]PARPi PET imaging was evaluated in orthotopic oral cancer models in comparison to [18F]FDG [26]. Clinically, FDG-PET is used to determine disease extent and post-therapy surveillance, which is complicated by physiological uptake in the head and neck region. The study showed that [18F]PARPi uptake was limited to tumor tissue and showed higher uptake in orthotopic tongue tumor xenografts compared to the surrounding tongue, which was not the case for [18F]FDG, indicating its feasibility for clinical applications in head and neck cancer imaging. Another study found that human papilloma virus (HPV)-positive and negative oropharyngeal cancer cells showed similar PARP1 expression and [18F]PARPi uptake, suggesting the tracer as an HPV-independent imaging tool for imaging in oropharyngeal cancer patients [36]. PARPi are also in clinical studies as mono- and combination treatments in small cell lung cancer (SCLC). In this context, [18F]PARPi imaging was introduced as a tool to measure the extent and duration of target engagement of the PARPi’s olaparib and talazoparib in patient-derived xenograft models of SCLC [38]. Since complete inhibition of PARP-mediated DNA repair is essential for efficient therapy, this real-time, non-invasive monitoring approach could be used to optimize dosing and timing of PARPi therapy [38]. A follow-up study later actually showed that treatment of SCLC PDX bearing mice with an efficacious and sub-efficacious dose of talazoparib (0.3 mg/kg vs. 0.1 mg/kg) correlated with [18F]PARPi uptake on PET imaging and treatment outcome [48]. [18F]PARPi was also evaluated as an alternative for [18F]FDG for diffuse large B-cell lymphoma (DLBCL) imaging in a syngeneic mouse model [39]. DLBCL treatment can induce inflammation, and [18F]FDG PET often faces difficulties in differentiating malignant from inflamed masses, e.g., in lymph nodes [39]. [18F]PARPi uptake was significantly higher in DLBCL lymph nodes compared to inflamed and normal lymph nodes, which also reflected PARP1 expression, while [18F]FDG uptake was similar in DLBCL and inflamed lymph nodes [39]. Lastly, [18F]PARPi was tested for its ability to distinguish recurrent tumor from radiation injury [40]. The inability to confidently distinguish these entities is an important clinical problem, e.g., in brain tumors, because it can result in delayed treatment decisions. In a mouse model of experimental radiation necrosis, [18F]PARPi showed no avidity to radiation injury (lesion/contralateral ratio: 1.02), while the brain tumor imaging tracer [18F]Fluorethyltyrosine showed increased lesion uptake (lesion/contralateral ratio: 2.12), indicating [18F]PARPi could be a more specific marker to discriminate these two pathologies non-invasively [40]. The Pimlott lab introduced [18F]20 as a PET imaging agent for glioblastoma in 2018. This tracer has a methylfluorobenzene instead of the cyclopropyl moiety of olaparib and is thus structurally very similar to [18F]PARPi with its fluorobenzene. Although it had a low IC50 value (1.3 nM in G7 human glioblastoma cells) and favorable tumor uptake (3.6% ID/g after 120 min), [18F]20 was eliminated from further investigations due to observed high skeletal uptake (8.5% ID/g) due to defluorination [49]. Both clinically translated PET imaging agents, [18F]FTT and [18F]PARPi, are characterized by hepatobiliary clearance, which complicates imaging of abdominal lesions, e.g., liver metastases. To address this limitation, Stotz et al. introduced [18F]FPyPARP as a less lipophilic variant by exchanging the fluorobenzoyl residue with a fluoronicotinoyl group [50]. A side-by-side in vivo comparison of [18F]FPyPARP to [18F]FTT and [18F]PARPi revealed a partial shift to renal clearance, but since tumor-to-liver ratios remained well below “1”, it is likely that further modifications and a stronger shift to renal clearance would be required for PARP1 imaging of abdominal lesions. In 2019, Wilson et al. (Cornelissen lab) reported [18F]olaparib, which is an isotopologue of olaparib, and hence the first directly radiolabeled PARPi without structural modifications [51]. Radiolabeling was achieved via a copper-mediated 18F-fluorodeboronation of a protected boronic pinacol ester precursor in a synthesis time of 135 min and an activity yield of 18% [51]. In vitro studies showed a correlation of [18F]olaparib uptake with PARP1 expression levels in pancreatic ductal adenocarcinoma (PDAC) cells. PET studies in mice bearing PSN-1 xenografts revealed a tumor specific uptake of 3.2% ID/g, which could be blocked. Furthermore, [18F]olaparib uptake increased by 70% after irradiation with 10 Gy, corresponding to an increase PARP1 expression in tumors [51]. Meanwhile, Guibbal et al. established procedures for automated radiosynthesis of [18F]olaparib (120 min, activity yield: 6%), which are compatible with Eckert and Ziegler Modular Lab systems, offering promising perspectives for production for clinical studies and routine use [52]. Furthermore, Bowden et al. was able to introduce a feasible automated copper-mediated radiofluorination, which led to an increase in activity yield (41%) and radiochemical yield (80%) [53]. Additional clinical and preclinical data obtained with [18F]olaparib are eagerly awaited by the scientific community.Since treatment resistance to olaparib is often associated with rapid elimination via drug transporters, especially p-glycoprotein (P-gp), AZD2461 was developed in 2016 as the next generation PARPi. AZD2461 showed similar anticancer potency as olaparib in vitro and in vivo but is a poor substrate for drug transporters. In addition, it showed lower levels of haematological toxicity in mice and was found to be a weaker inhibitor of PARP3 than olaparib [54]. Evading P-gp drug transport should also lead to better penetration of the BBB. To test this hypothesis and investigate the role of PARP1 in neuroinflammation and neurodegenerative diseases, Reilly et al. (Mach lab) developed an 18F labelled analogue of AZD2461 in 2019, called [18F]9e. However, [18F]9e showed non-appreciable brain-uptake in non-human primates, suggesting that [18F]9e does not cross the BBB and is hence not suitable to investigate PARP1 in neurodegenerative diseases [55]. Almost in parallel, a radiofluorinated isotopologue of AZD2461 was synthesized via copper-mediated 18F-fluorodeboronation (Gouverneur and Cornelissen lab). [18F]AZD2461 was evaluated in pancreatic cancer cell lines and a xenograft mouse model of pancreatic cancer in comparison to [18F]olaparib, which was developed in the same lab [56]. Cellular uptake of [18F]AZD2461 in PSN-1 cells was less than 50% compared to the [18F]olaparib. Interestingly, blocking with olaparib or AZD2461 only reduced the [18F]AZD2461 uptake to 70% and 25% of the initial binding, respectively, while both olaparib and AZD2461, could completely block [18F]olaparib uptake. In vivo, [18F]AZD2461 uptake could also not be blocked completely, but curiously olaparib was more efficient at blocking than AZD2461 [56]. These results could suggest that AZD2461 has other, currently unknown, targets in addition to PARP1 and PARP2 and is hence less suitable as a PARP imaging agent. In parallel to 18F-labelled olaparib analogues, iodinated derivatives based on the same 2H-phthalazin-1-one scaffold of [18F]PARPi were developed, since the variety of iodine isotopes could enable imaging with PET (e.g., 124I and SPECT (e.g., 131I), but also radionuclide therapy (e.g., 131I, 123I and 125I). Here, it needs to be considered that the large molecular weight of iodine could negatively affect the molecule’s membrane penetration capability and pharmacokinetics and hence, tumor uptake. In 2015, Salinas et al. synthesized a series of meta and para-iodinated olaparib analogues with different linker lengths between the aromatic ring and the olaparib core, resulting in compounds with IC50-values between 9 and 107 nM. The group identified [124/131I]I2-PARPi (para-iodinated) as the lead candidate, which showed high PARP1 affinity (IC50 = 9 nM) and specificity, shown by blocking. In vivo, [124/131I]I2-PARPi was able to delineate orthotopic glioblastoma xenografts using PET as well as SPECT imaging and yielded tumor-to-brain ratios of 40 ± 6.3 in U251 MG xenografts 2 h p.i. [57]. Simultaneously, Zmuda et al. reported in 2015 an 123I-labelled version of I2-PARPi using the same precursor and coupling conditions as Salinas et al., called [123I]5 [58]. This radiotracer was evaluated as a potential SPECT imaging agent for glioblastoma as well and reached tumor-to-muscle ratios of 5.6 ± 2 at 2 h p.i. in a subcutaneous U87 MG model [58].One group also evaluated 11C as possible radionuclide to create [11C]olaparib. However, the work of Andersen et al. showed that the palladium complexes which were used as a precursor for the labeling reaction were unstable [59]. Despite a continuing effort to develop optimized reaction conditions for the 11C-labeling reaction [59,60], fast progress with 18F-labeling and its longer half-life led to a stronger focus on 18F-labelled PARP inhibitors. Reporting about “exotic” radionuclides, the work of Huang et al. needs to be mentioned. Therefore, [64Cu]DOTA-PARPi with a 64Cu-chelating system was conjugated to the olaparib precursor. Unfortunately, although tumor uptake in mesothelioma mice models reached 3.45% ID/g after 1 h, the conjugation of the DOTA-chelating moiety led to a decrease in binding affinity by 40 [61]. These findings underline that such large structural modifications compared to the parent PARPi cannot be tolerated in the design of PARP imaging agents. The rucaparib scaffold was developed in 2008 by a collaboration between the University of Newcastle and Agouron Pharmaceuticals [62]. Menear et al. followed up the development and discovered the inhibitory potential of rucaparib towards PARP [41], which led its first phase I clinical study in combination with temozolomide in patients with advanced, solid tumors [63]. Ten years later, in 2018, the EMA approved rucaparib to be used in patients with HR deficient ovarian cancer [64]. Now, it is also FDA-approved for the treatment of HR deficient metastatic castration-resistant prostate cancer [65]. In parallel, several radiolabeled rucaparib analogues were developed. Zhou et al. (Mach lab) developed the first 18F-labelled radiotracer structurally closely related to rucaparib, [18F]F12, later called [18F]Fluorthanatrace, in 2014 [66]. [18F]Fluorthanatrace (short: [18F]FTT) was derived from AG14361 [67], not AG014699/rucaparib, by replacing the dimethyl phenylmethanamine with 18F-fluoroethoxy benzene [62]. AG14361 was a former candidate for clinical development by Agouron, but later rucaparib was chosen due to better in vitro potency and in vivo efficacy [62,63]. Structurally, rucaparib features a fluorination and possesses an amine group on the indole moiety, which are absent in AG14361. [18F]FTT displayed a good affinity (IC50 = 6.3 nM) towards PARP1 and showed specific tumor uptake (3–5% ID/g 1 h p.i.) in MDA-MB-231 and MDA-MB-436 xenograft models [66]. In a panel of breast cancer cell lines, [18F]FTT uptake was compared in BRCA-mutant HCC1937 (high PARP1 expression) to the BRCA-wildtype MDA-MB-231 and MCF-7 cells and corresponded to these different expression levels [68]. In vivo imaging showed the best tumor delineation in the HCC1937 xenografts, with tumor-to-muscle ratios of 1.9 [68]. Another study from the same lab also found higher [18F]FTT uptake in BRCA-mutant (SNU-251) than BRCA-WT (SCOV3) cells, corresponding to protein expression levels. Since expression and radiotracer uptake was higher in the BRCA-mutant cell line, corresponding to its higher sensitivity to PARPi treatment and radiation, the authors suggested that [18F]FTT could be used to predict treatment efficacy [69]. The same approach was investigated in a study focused on ovarian cancer. Here, it was first shown that PARP1 knockout cells and mice showed resistance to PARPi treatment, confirming that PARP1 expression is a requirement for PARPi sensitivity. The authors observed that [18F]FTT tumor uptake was decreased in olaparib treated animals compared to untreated animals, concluding that the tracer is suitable to measure PARP1 expression in vivo [70].In 2018, Zhou et al. developed a modified version of [18F]FTT, called [18F]WC-DZ-F [71]. The radiotracer was radio-fluorinated directly at the para position of the benzene ring, in exchange for the fluoroethoxy group of [18F]FTT. This compound was characterized in a subcutaneous prostate cancer model, where tumor uptake was around 4% ID/g 2h p.i. [71]. Although [18F]WC-DZ-F showed a higher in vivo blood stability compared to [18F]FTT (78.5% vs. 13% at 30 min), substantial nonspecific uptake in bone and muscle were observed in the biodistribution data, limiting the potential advantages over [18F]FTT [71].Recently, the first 18F-radioisotopologue of rucaparib was developed (Cornelisson and Gouverneur labs) using a synthesis strategy involving Cu(II)-mediated 18F-fluorodeboronation followed by reductive amination, to obtain [18F]rucaparib where the fluorination took place at the aromatic ring system of the benzimidazole core [72]. Similar to olaparib/[18F]olaparib, [18F]rucaparib is expected to have identical properties and pharmacokinetics as its parent molecule. The first in vivo imaging data with [18F]rucaparib are eagerly awaited. In addition to the extensive research efforts with regards to olaparib and rucaparib-like radiotracers, a few PARPi based on other natural structures have been developed. In 2005, before the discovery of rucaparib and olaparib as PARPi, Tu et al. were working on the very first example of a PET tracer targeting PARP [73]. [11C]PJ34 was a phenanthridinone derivative, which was able to block NAD+ from its natural binding site on the PARP enzyme. Importantly, hyperactivation of PARP leads to the depletion of NAD+ inside the cells, which can induce necrosis or lead to related diseases, such as diabetes [73]. Using streptozotocin-treated rats (type I diabetes model), a high uptake of [11C]PJ34 in the target organs, the liver and pancreas, was observed. This indicated the possibility of [11C]PJ34 to target PARP during its hyperactivation, which is a key driving mechanism for necrosis-related diseases [73]. However, further studies with this radiotracer were not conducted. While all other PARP imaging approaches are based on radiolabeled PARPi, Shuhendler et al. pursued a different approach and developed a radiofluorinated NAD+ analogue with the goal to image parylation activity instead of PARP expression [74], since PARP activity could be a better predictor for PARPi therapy response. Indeed, [18F]SuPAR showed increased tumor uptake in HeLa and MDA-MB-231 xenografts after radiation treatment, which significantly correlated with increased PAR levels after the DNA damage inducing treatment. The specificity of [18F]SuPAR was shown by a decreased tumor uptake after blocking with the PARPi talazoparib in mice. Despite these promising results, it should be noted that in vivo assessment of PARP activity was complicated by the fact that NAD+ also serves as substrate for other enzymes and plays important roles in enzyme catalyzing redox reactions.In addition, the first talazoparib-based radiotracers have recently emerged. Talazoparib (IC50 = 0.6 nM) is a PARPi with a similar affinity (IC50 = 0.6 nM), but much higher potency than olaparib and rucaparib (IC50 = 1.9 nM and 2.0 nM, respectively), which is often attributed to its high PARP-trapping capacity and its broader selectivity profile [18,75]. Of note, talazoparib is given clinically at much lower daily doses (1 mg/day) than olaparib and rucaparib (300 mg twice daily), due to its higher potency and toxicity. Talazoparib was approved by the FDA (2018) and EMA (2019) for the treatment of germline BRCA-mutated, HER2-negative metastatic breast cancer [16]. It is further clinically tested, e.g., in metastatic breast cancer patients with a deleterious somatic BRCA mutation and in men with DNA repair defects additional to their metastatic castration-resistant prostate cancer. Two research groups reported the radiosynthesis of [18F]talazoparib isotopologues in 2021 using different strategies. Zhou et al. (Katzenellenbogen and Xu lab) largely followed procedures in line with the original non-radioactive synthesis of talazoparib [76] and pursued early stage 18F incorporation [77], while Bowden et al. (Maurer lab) established a late stage 18F incorporation route to obtain the radiotracer [78]. Bowden et al. achieved automated radiosynthesis of [18F]talazoparib, yielding an enantiomerically pure compound. This is important, since talazoparib possesses two distinct chiral centers, of which only the (8S, 9R)-diastereomer is a potent PARPi [78]. Subsequent in vitro experiments showed a blockable radiotracer uptake of ~22% of added activity in HCC1937 cells compared to the less potent (8R, 9S)-diastereomer with ~0.3% uptake in the same cell line [78]. In vivo biodistribution data in HCC1937 xenograft-bearing mice showed a tumor uptake of 3.7 ± 0.7% ID/g, but tumor-to-muscle ratios of only 1.8 ± 0.4 at 2.5 h p.i. [78]. Zhou et al. (Katzenellenbogen and Xu lab) also synthesized [18F]talazoparib with high chiral purity in an alternative synthetic route, involving less steps and different fluorination conditions compared to [78]. [18F]talazoparib showed high tumor uptake in in PC-3 prostate cancer xenografts, which slightly increased from 4 h (3.8 ± 0.6% ID/g) to 8 h p.i. (4.5 ± 0.3% ID/g) [77]. The biodistribution was rather similar to Bowden et al. and showed high uptake in liver, spleen, kidney, and pancreas that only slightly reduced over time. Imaging data are not reported in this study. Both studies indicate that [18F]talazoparib shows slower washout from organs than olaparib and rucaparib-based radiotracers, with high organ uptake in the spleen, liver, and kidneys, which could be challenging for imaging applications with 18F. Nevertheless, Zhou et al. suggest that the prolonged tumor retention could be an advantage for radiotherapy applications, which could be studied using bromo- and iodo-derivates reported in the same publication [77]. In addition to the large body of preclinical work, which investigated a variety of potential clinical scenarios, results from two clinical studies centered on [18F]PARPi PET imaging were published to date, both in 2021 (Table 1). The first-in-human trial of [18F]PARPi investigated safety and feasibility of PET/CT imaging in head and neck cancer patients (NCT03631017) [79]. PET/CT scans and analysis of blood samples of 11 patients with oral and oropharyngeal cancer were obtained 30, 60, and 120 min post injection. The patients received an [18F]FDG scan as well, which was compared to [18F]PARPi. The tracer was well tolerated by all patients with only one patient experiencing grade 1 mucositis. All primary tumors (n = 10) and FDG-avid lymph nodes (n = 34) could be visualized with [18F]PARPi with an average SUVmax of 2.8 ± 1.2 at the 120 min timepoint, which yielded the highest lesion-to-background contrast. Rapid clearance of [18F]PARPi from healthy organs was observed between the 30 and 120 min timepoints, whereas the activity persisted longer in primary tumors and the metastatic lymph nodes [79]. The study reports that [18F]FDG uptake yielded, on average, higher SUVmax values in tumors and metastatic lymph nodes, but [18F]PARPi uptake was less variable. Furthermore, the authors report that [18F]PARPi imaging resulted in an average dose of 3.9–5.2 mSv per scan, which is lower than an average FDG scan (8.1 ± 1.2 mSv) [80]. Interestingly, on top of all FDG-avid lesions, [18F]PARPi showed uptake in six additional lymph nodes. However, the phase I study protocol did not allow the biopsy of these lesions or conduct general histological confirmation of the imaging results. In this study, patients received on average 290 pmol [18F]PARPi, which is 6.7 orders of magnitude lower than a typical daily dose of olaparib (2 × 300 mg) during an active treatment cycle [79]. In the second clinical study of [18F]PARPi, which was focused on brain cancer (NCT04173104), PET/MR imaging of five brain cancer patients was conducted [43]. The tracer showed higher uptake in active brain tumor lesions (SUVmean = 1.16) compared to regions associated with treatment-related changes (SUVmean = 0.45) at 60 min p.i. and tracer uptake could be correlated with PARP1 expression via immunohistochemistry [43]. Although only an anecdotal observation, heterogeneity in intratumoral [18F]PARPi uptake in one patient could be connected to areas of high and low PARP1 expression in histological analysis (Figure 4A). Overall, the study indicates uptake specificity, the ability to cross the BBB, and confirms the very low uptake of [18F]PARPi in normal brain tissue, which is promising for brain cancer imaging, but larger patient cohorts are needed to confirm these results.Selected clinical PET imaging results of [18F]PARPi and [18F]FTT. (A) PET/MR imaging of [18F]PARPi (NCT04173104) in a brain cancer patient showed heterogenous uptake, which corresponded to areas of higher and lower PARP1 expression in histological analysis [43]. (B) [18F]FTT imaging of an ovarian cancer patient (NCT02637934) showed clear tumor visualization (green arrow) and delineation (SUVmax = 5 g/mL) and absence of bladder uptake observed with [18F]FDG PET (yellow arrow) [70]. (C) [18F]FTT uptake in breast cancer patients (NCT03846167). Subject 1 had clear tumor uptake pretherapy (SUVmax breast 4.7 g/mL) and a blockade of uptake posttherapy (SUVmax breast 2.4 g/mL) and went on to have a response to PARPi. Subject 2 had minimal uptake pretherapy (SUVmax breast 2.3 g/mL) and a similar uptake posttherapy (SUVmax breast 2.4 g/mL) and had progression on PARPi [81]. Copyright notice: (A) Reprinted with permission from [43], Copyright, 2021, Society of Neuro-Oncology. (B) Reprinted with permission from [70], Copyright, 2018, American Society for Clinical Investigation. (C) Reprinted from [81] under Creative Commons CC BY 4.0.Overview of all currently ongoing or finished clinical trials of introduced radiotracers.1 At Memorial Sloan Kettering Cancer Center in New York, United States. 2 At Washington University School of Medicine in Missouri, United States. 3 At Abramson Cancer Center of the University of Pennsylvania in Pennsylvania, United States. 4 At University of Pennsylvania in Pennsylvania, United States. 5 At National Cancer Institute in Texas, United States. 6 At MD Anderson Cancer Center, Houston, Texas, United States.[18F]FTT is the PARP imaging agent with the most extensive clinical evaluation to date. At the time of this writing, nine studies are registered in clinicaltrials.gov (keyword: FluorThanatrace; accessed 11 February 2022). Four of them are early phase 1 (phase 0), three are phase 1 trials, one was not assigned a phase, and one is phase 2. The first clinical data, published in 2017 (NCT02469129), included a small cohort of eight patients with different malignancies and provided the first evidence that clinical [18F]FTT imaging is feasible [82]. PET images from this study showed visible [18F]FTT uptake in tumor regions from one out five patients with measurable tumors who had a biphenotypic hepatocellular carcinoma/cholangiocarcinoma [82]. An erratum to the study corrected that a patient with pancreatic ductal adenocarcinoma, who was originally reported to show [18F]FTT uptake, did not demonstrate [18F]FTT uptake above the background activity [83]. The effective dose was estimated at 6.9 mSv, which is in a similar range of a [18F]FDG PET scan.In 2018, Makvandi et al. reported results from NCT02637934, where 18 patients with epithelial ovarian cancer types underwent [18F]FTT and [18F]FDG PET/CT imaging (Figure 4B) [70]. Researchers observed [18F]FTT uptake in patients using PET/CT with maximum standardized uptake values (SUVmax) ranging from 2–12 g/mL (clear delineation of tumor region for SUVmax > 5 g/mL). Further correlation of PET imaging with PARP1 immunofluorescence staining and autoradiography was found, but not with [18F]FDG imaging [70]. Recently, more data from this trial were published, with a special focus on the pharmacokinetics of [18F]FTT [84]. Data from 14 patients over the course of 60, 90, and 180 min post-injection, were analyzed. For the 0–60 min dynamic scan time points, the kinetic parameters (e.g., 2-tissue-compartment model with reversible binding) and SUVmax values were in correlation with PARP immunofluorescence data (r = 0.80 and r = 0.93, respectively). Stability of the radiotracer after 60 min was confirmed via computational kinetic analysis showing 59% of parent [18F]FTT was still intact in pooled plasma samples. Interestingly, at longer dynamic scan times of t = 110 min and 199 min, the tumor uptake increased, suggesting a possible irreversible binding (model) as a consequence of PARP trapping [84]. Published results are also available from NCT03083288 [85] and NCT03846167 [81]. Both studies evaluated [18F]FTT for the quantification of PARP expression levels in breast cancer patients using PET/CT imaging. In NCT03083288, 30 breast cancer patients (stage I to IV) with a range of breast cancer phenotypes (estrogen receptor-positive, human epidermal growth factor receptor-positive or triple negative) were enrolled and the BRCA status was analyzed. The study showed that [18F]FTT uptake was highly variable among the different subtypes of breast cancer and showed similar variability within each subtype (SUVmax = 2.6–11.3 g/mL). Furthermore, patients with and without BRCA1/2 mutations had a similar range of tumor uptake levels (SUVmax = 2.9–11.3 g/mL) [85]. This is interesting, since BRCA1/2 status is currently used as only a biomarker for PARPi treatment eligibility, but the response patterns are still not well understood. Potentially, varying levels of PARP expression could also contribute to the treatment response within the eligible patient population. Clinical studies involving pre-treatment PARP-PET and correlating the uptake to the treatment response would be required to answer this question. In NCT03846167, four breast cancer patients with invasive ductal carcinoma (stage III/IV, 3 triple-negative and 1 estrogen receptor-positive) underwent [18F]FTT PET/CT imaging pre- and one week post-PARPi treatment (Figure 4C) [81]. Within this group, three patients had moderate [18F]FTT uptake pre-treatment (SUVmax range: 4.2–6.8 g/mL) and subsequently showed stable disease or tumor regression in response to PARPi treatment. The fourth patient did not show [18F]FTT uptake above background in any lesion pre-treatment and also did not respond to PARPi therapy. The study also found that [18F]FTT uptake was reduced to background levels in all patients in the “post treatment” scan [81]. Although not clearly stated, we assume that PARPi treatment was still ongoing at the time of the second scan in order to show PARPi target engagement. Despite the small number of patients, these results are promising and warrant further studies into prediction of treatment response and measurement of target engagement in clinical studies. PARP-targeted radiotherapy offers the exciting prospect of delivering cytotoxic radiation directly to the tumor cell nucleus, and therefore, the DNA, instead of the cell membrane or tumor microenvironment, raising hopes to more efficiently introduce DNA damage compared to extranuclear radioligand therapy agents. Especially the use of radioisotopes with strong linear energy transfer (LET) and short path lengths, i.e., Auger electron emitters like 123I (t1/2 = 13.2 h [86]) or 125I (t1/2 = 60 d [87]) and α-emitters, such as 211At (t1/2 = 7.2 h [88]) could find a highly effective application using PARPi as intranuclear delivery vehicles [89,90]. Auger electrons have a high LET of 4–26 keV/µm with an extremely short tissue range of 2–500 nm, which means that they only cause lethal damage when emitted in direct vicinity to sensitive structures, such as the DNA or the cell membrane [89,91]. A-emitters have an even higher LET of about 80 keV/µm and a moderate pathlength of 50–100 µm, which covers up to 5 cell diameters. Some studies have reports that only 1–10 α-particle traversals are necessary to effectively kill a target cell [92]. Β-emitters, such as 131I or 177Lu, have a greater path length of up to 1–12 mm with an LET of 0.2 keV/µm, meaning they can be suitable to treat larger tumor masses but can also cause damage in adjacent tissues and organs [89]. Key characteristics of the therapeutic radioisotopes that have been conjugated to PARPi to date are summarized in Table 2 and main parameters and outcomes from preclinical in vivo studies are summarized in Table 3. The therapeutic efficacy of the β-emitting radiotracer [131I]PARPi ([131I]I1-PARPi from [57]), was evaluated in mice bearing U251 MG or U87-p53 human glioblastoma xenografts [96]. Specific uptake and tumor retention after intratumoral injection of the tracer was shown in two ways–the tracer could block uptake of [18F]PARPi and uptake of [131I]PARPi could be blocked by olaparib [96]. To assess the therapeutic property of [131I]PARPi, mice bearing U87-p53 tumor cells were assigned to three different cohorts: a control group (treated with PBS) and two fractionated treatment groups with either the therapeutic agent [131I]PARPi (3 × 14.8 MBq) or its non-radioactive version [127I]PARPi [96]. A median survival of 29 days was observed in the treatment group with [131I]PARPi while the control group and the group with the non-therapeutic [127I]PARPi had a lower median survival (22 and 20 days, respectively) (Figure 5A) [96]. The Auger-emitting version of the same compound, [123I]MAPi, showed a 16-fold greater cell killing potency compared to [131I]PARPi (EC50 = 69 nM and 1148 nM, respectively) and induced higher levels of DNA damage [97]. After intratumoral injection, the tracer was retained in tumors at high levels (40% ID/g 18 h p.i.), which could be blocked with a systemic olaparib pre-injection, and uptake in other organs remained low. Treatment of TS543 tumor-bearing animals with a single intratumoral dose of [123I]MAPi (0.37–1.11 MBq) led to an increase in survival (58 days vs. 40 days in the control group). These results were confirmed in a second cohort, where the radiotherapeutic was delivered via an osmotic delivery pump over a prolonged time. In this setup, survival increased from 48 days in the control group to 72 days in the treated group and the treatment was tolerated well by the animals [97]. Subsequently, an improved synthesis route for the 123I-labeling was developed using a single step 123I-iododestannylation reaction, yielding higher molar activity (Am) of 11.8 GBq/µmol compared to previous work with Am = 3.9 GBq/µmol [98]. While intratumoral injection might be considered for glioblastoma treatment clinically, for other tumor entities, it is not an option. Therefore, the therapeutic potential of [123I]MAPi after systemic injection was evaluated in a colorectal cancer model comparing p53+/+ to p53−/− models [99]. In tumors, PARP1 expression is elevated in the nucleus, increasing the likelihood for the Auger-emitting isotope to induce DNA damage. The combination of high PARP1 expression and genomic instability in tumors, e.g., via a p53 loss, could explain the promising therapeutic efficacy and tolerable toxicity of [123I]MAPi after systemic application, which needs to be confirmed in further studies.Overview of in vivo efficacy studies using PARP-targeted radiotherapeutics.Control1 × 36 MBq/kg (~720 kBq) i.v.3 × 200 µg PD-L1 i.p.Combination1 day21 days38 days65 daysControl1 × 555 kBq i.v.1 × 1110 kBq i.v.1 × 1480 kBq i.v.Fractionated (370 kBq i.v. 2 × weekly)35 days61 days65 days10 days (toxicity)80 daysVehicle PBS i.t.1 × 9.9 nmol [127I]PARPi i.t.3 × 14.8 MBq [131I]PARPi i.t.22 days20 days29 daysVehicle i.t.1 × 0.37–1.11 MBq [123I]MAPi i.t.Vehicle osmotic pump delivery[123I]MAPi osmotic pump delivery40 days58 days48 days72 daysVehicle5 × 74 MBq [123I]MAPi i.v.3.429 weeks3.286 weeksVehicle i.v.5 × 80 µg/kg [127I]MAPi i.v.5 × 74 MBq [123I]MAPi i.v.2.429 weeks3.071 weeks3.714 weeks* PFI: Progression free interval.For therapy, mice were administered 5 cycles of up to 74 MBq [123I]MAPi, which led to an increase in median survival in the [123I]MAPi treated group in HCT116 p53−/− animals (3.2 weeks) compared to the vehicle treated controls (2.4 weeks) (Figure 5B), but not in HCT115 p53+/+ animals ([123I]MAPi: 3.3 weeks, vehicle: 3.4 weeks), supporting that loss of the tumor suppressor p53 lead to increased sensitivity. Although biodistribution data showed that large fractions of [123I]MAPi pass through the hepatobiliary system and uptake in several organs was higher than in tumors, only minimal systemic toxicity was observed in a toxicity study after GMP guidelines. It is hypothesized that during excretion, metabolism confines the agent to the perinuclear region of the cell and therefore puts it outside the range of an Auger-emitter to achieve significant damage upon cellular DNA [99]. Recently, another therapeutic study of an Auger-emitting PARPi, [125I]PARPi-01 (isotopologue of [131I]I2-PARPi from [57]), was published (Morgenroth Lab) [102]. To assess the theranostic efficacy of the Auger electron emitter on triple negative breast cancer (TNBC), the tracer was evaluated in 11 different TNBC tumor cell lines, including BRCA-mutated and BRCA-wt cell lines. Specifically, [125I]PARPi-01 uptake was shown via olaparib blocking in MDA-MB-231 cells [102]. While some cell lines already showed a sensitive response to [125I]PARPi-01 monotherapy at lower concentrations than olaparib, the therapeutic response could be improved in non-responsive cell lines using a combinatorial treatment of [125I]PARPi-01 with the chemotherapeutic drug Dox-NP. The observed responses were consistent across a panel of assays including cell cycle analysis, apoptosis quantification, and colony formation assays [102]. If these results could be confirmed in vivo, [125I]PARPi-01 could be another interesting candidate for PARP radiotherapy.The lab of Robert Mach developed an 125I-labelled PARP1-targeted tracer, based on the same scaffold as [18F]FTT [103]. [125I]KX1 showed uptake in HCC1937 and MDA-MB-231 xenograft models, where a tumor uptake of 5% ID/g 2 h p.i. (HCC1937) and 3% ID/g 2 h p.i. (MDA-MB-231) was observed [103]. These data aligned with the known PARP expression levels of these cell lines and were also confirmed by ex vivo autoradiography of the tumors of both cell lines. However, unlike [18F]FTT, tumor uptake could not be significantly blocked by pre-injection of olaparib [103]. Further, [125/123I]KX1 was tested in ovarian cancer cells and human ovarian cancer xenograft mouse models [104]. In vitro experiments showed PARP1-dependence of the cell killing effect and a dose-dependent increase in the number of γH2AX foci after treatment with [125I]KX1 [104]. The authors also showed a dose-depending increase of apoptosis on tumor slices from patients upon [125I]KX1 treatment [104]. [125I]KX1 was evaluated for treatment in neuroblastoma models where its cytotoxicity was 104–106 times higher than its non-radioactive precursor KX1 across a panel of 19 cell lines [87]. In this study, an α-emitting KX1 version, [211At]MM4, was presented and showed significantly higher cell-killing potential than [125I]KX1, indicating that much lower doses would be needed to induce therapeutic effects. In vivo tumor dosimetry confirmed the superior therapeutic properties of [211At]MM4 over [125I]KX1, yielding a 150-times higher tumor nuclei dose per decay (radiation dose of ~ 35 cGy/decay vs. 0.1 cGy/decay, respectively) (Figure 5C) [87]. Hence, [125I]KX1 would require significantly higher activity than [211At]MM4 for equivalent in vivo efficacy. In combination with the long half-life of 125I, this could limit its in vivo potential. Moreover, immunohistochemistry confirmed that [211At]MM4 caused dose-dependent DNA damage among neuroblastoma cell lines, resulting in an increase of ƴH2AX foci [101]. Additionally, comparing the sensitivity of UWB1.289 (BRCA1 deficient) and UWB1.289-BRCA1 restored cells towards [211At]MM4, no difference was found, suggesting the therapeutic effect of 211At does not depend on the HR status of the cell line [101]. Lastly, a therapy study in IMR-05 tumor-bearing mice showed significantly increased median survival in the treatment groups (555 kBq and 1110 kBq single dose of intravenous [211At]MM4) over the control group (61 and 65 days vs. 35 days, respectively) [101]. In addition, the animals tolerated the treatment well and showed no weight loss or other signs of systemic toxicity, rendering [211At]MM4 a promising candidate for further evaluation as radiotherapeutic PARPi. In addition to inhibiting PARP, it is also feasible to inhibit other tumor escape pathways. One such method is the blocking of the PD-1 immune-checkpoint, which is normally used by the tumor cells to evade the tumor surveillance mechanism of the body [100]. In order to enhance the immune-checkpoint blockade, the α-emitter [211At]MM4 was tested in mono- and combination therapy on mice bearing GL26 glioblastoma tumor cells (Figure 5D) [100]. Hereby, average tumor response was the greatest (100%) for the combination treatment compared to the mono treatments with either 200 µg anti-PD-1 (83.6%) or 36 MBq/kg [211At]MM4 (58.2 %) [100]. Similar results were observed for the average progression free intervals (65, 36.4 and 23.2 days, respectively) and for the percentages of disease-free mice at the end of the study (100%, 60% and 0%, respectively), suggesting [211At]MM4 to be a potential candidate for combinatorial therapy of glioblastoma with PD-1 immune-checkpoint blockade [100].A second 125I-labelled rucaparib analogue was reported, namely [125I]KX-02-019 [105]. This compound is a modified version of [125I]KX1 that features a bicyclical benzimidazole. Its Ki value was in favorable range (13.9 nM) and biodistribution studies with mice bearing EMT-6 tumors showed tumor uptake of 1% ID/g at 2 h p.i. [105]. Although this was lower than other PARP imaging agents, tumor-to-muscle ratios were about five and partial deiodination in vivo was found due to thyroid uptake. Surprisingly, this study revealed higher affinity of the radiotracer towards PARP2, and therefore may be useful to predict treatment response to PARPi therapy more precisely [105].To briefly compare the results observed with both therapeutic isotopes 125I and 211At, it can be stated that, although based on the microdosimetry of the nuclides, 125I should be more effective than 211At at cell killing when bound or in very close proximity to the DNA; PARP-targeted 211At therapy was much more potent than 125I therapy. Of the many radiolabeled PARP imaging and therapeutic agents, only two are currently on clinical trials: [18F]PARPi and [18F]FTT. Next to the general suitability of a PARP-targeted tracer for translational/clinical imaging, several factors can hinder the progression of radiopharmaceuticals from the preclinical to the clinical phase and large scale clinical routine production, including choice of chemicals, feasibility of automation, total synthesis time, and final dose achievable [106,107]. In the following, we will outline challenges for PARP imaging agents related to synthesis automation, upscaling, and materials suitable for human use.Automation. The vast majority of radiopharmaceuticals with clinical applications are produced on an automated synthesis platform. Exceptions can be found, most commonly among compounds labelled with radiometals. These radiochemicals typically require fewer steps and no purification, making manual synthesis possible without exposing the operator to high levels of radiation. However, in the vast majority of cases, and especially with 18F and 11C, radiosynthesis automation is a fundamental step towards the clinical validation of a radiopharmaceutical. Full automation allows radiopharmacies to start the synthesis at a much higher radiation levels than manual processes, as the operators set up the equipment before delivering the activity in the hot cell and are therefore protected from radiation exposure. Additionally, a fully automated system ensures synthesis reproducibility, with parameters, including pressure, time, and temperature, being finely controlled. Finally, the most likely source of microbiological contamination in a GMP laboratory is a human operator; thus, limiting manual intervention reduces the risk of contaminating the final product [108]. For some of the PARP radiopharmaceuticals discussed in this review, namely the BODIPY analogue of [18F]PARPi, [18F]PARPi-FL, the 18F-fluoroethyl analogue of AZD2461, the 18F-fluorethyl version of rucaparib, [18F]SuPAR, [18F]olaparib, and [18F]AZD2461, an automated procedure was already published [34,52,55,56,66,74]. For many, however, only a manual radiosynthesis was performed, and in some cases the reported procedures might be difficult to perform on an automated platform. Implementation on cassette-based systems, for example, need to consider limitations caused by dead volumes; therefore, processes that use very small volumes will likely need to undergo a series of additional tests to adjust. Furthermore, development of automated processes for particularly long and convoluted manual syntheses might prove challenging because of hardware limitations. Most systems have a limited number of positions available; therefore, processes requiring extra steps, such as multiple filtrations or solid phase extractions, will struggle to be accommodated on a smaller system [71,72,77]. Few systems have more than one heating reactors and/or magnetic stirrers; therefore, processes requiring these additional components will likely need re-examination, depending on the equipment available on site [72]. Typically, only one column can be connected to an automation system to perform the purification step; purifications requiring more than one column would most likely need to be revised [77]. Finally, processes requiring some sort of solid support, which must be removed at a subsequent step, will similarly need revision for successful implementation on an automated system. The synthesis of [18F]AZD2281, for example, requires magnetic removal of the beads used for the labeling [32,58]. Scale-up. As operators have to intervene heavily during the synthesis, manual and semi-automated radiosynthesis procedures are mostly carried out starting with smaller radiation levels than the amounts typically used for clinical applications.Radiosynthesis scale-up, especially of radiopharmaceuticals made with short-lived isotopes, has several benefits, including increased final dose, and therefore is able to scan more than one patient with a single synthesis and transfer of doses to centers at certain distances from the production site. For instance, while the starting activity of manually synthesized 18F PET tracers herein reviewed was reported, it was, in most cases, in the 500–1800 MBq range, and it could be increased to 20–30 GBq for some automated syntheses [55,56]. However, there are also certain challenges involved in scaling-up radiosynthesis. Decomposition of the reformulated product due to radiolysis, which is more likely at high starting activity, is the major concern; this occurrence is caused by the formation of highly reactive species (hydroxyl radicals, aqueous electron, and superoxide) from water, and can be mitigated by adding anti-oxidants; for instance, radiolysis of [68Ga]-NOTA-sdAb was prevented for up to 3 hours with the combination of 20% ethanol and 5 mg ascorbic acid [109]. Occurrence of radiolysis of the final product experienced during the scale-up phase have been previously reported. The radiochemical purity of [18F]AV-19, a PET tracer for amyloid plaques, decreased to 73% when the synthesis was scaled up to 66.6 GBq of 18F due to the decomposition of the product into four polar radioactive species [110]. Radiolysis issues were not considered in the PARP imaging literature, presumably because no large-scale production was reported to date. Small-scale radiosynthesis methods have been improving, including droplet radiochemistry and microfluidics, and show many advantages, such as lower costs and shorter processes; however, they are more suitable for preclinical use, since only small amounts of radiopharmaceuticals are produced [111]. Therefore, PARP-targeted PET tracers currently at the preclinical evaluation stage will likely need to undergo scaling-up, ensuring that the radiochemical yield is not negatively affected, and that the product is not subject to radiolysis, before moving to clinical tests.Materials. Pharmaceuticals for human use must be declared safe for the aforementioned purpose; one of the requirements towards this goal is to prove that any impurity in the final formulation is within the permitted daily exposure (PDE) for the specific administration route. The ICH provides guidelines on the PDE values of various types of impurities [112]. Of particular interest are byproducts, residual solvents, and elemental impurities. While radiopharmaceuticals usually go through several steps, such as solid phase extraction and column purification, that could remove these unwanted components, proving that the PDE values are within the limits is still necessary.Solvents are classified by the ICH guidelines into four groups. Class I solvents are to be avoided, either because they are known or suspected human carcinogens, or because of environmental concerns. Class II solvents should be limited in pharmaceutical products because of their inherent toxicity. Class III solvents have low toxic potential. Class IV includes solvents for which no adequate toxicological data was found. Most radiolabeled PARP agents in this review are synthesized in solvents already classified by the ICH guidelines, and for some of which the PDE is reported; notable examples are: (i) acetonitrile (class II, PDE = 4.1 mg/day) [32,68,71], occasionally mixed with other solvents (tert-butyl alcohol, class II, PDE = 35 mg/day [58], used for [123I]I-PARPi), or methanol, class II, PDE = 30 mg/day [97], used for [123I]MAPi); (ii) N,N-dimethylformamide (class II, PDE = 8.8 mg/day), used for [18F]FTT [66] and (iii) dimethyl sulfoxide (class III, PDE = 50 mg/day), used, among others, for [18F]SuPAR and [18F]PARPi [42,55,74,96]. Some, however, are synthesized in solvents that are not included in any of the four groups in the ICH guidelines, and do not have readily available toxicological data. For instance, [18F]olaparib, [18F]AZD2461, and [18F]rucaparib are synthesized in 1,3-dimethyl-2-imidazolidinone. In such cases, the clinical validation radiochemist will likely need to re-optimize the radiosynthesis in a different solvent [52,56,72]. Additionally, inclusion in the ICH guidelines does not imply that a standard test is described in a Pharmacopoeia; in such cases, in-house methods must be developed. Innovative radiosynthetic approaches have vastly increased the number and nature of molecules that can be potentially radiolabeled. However, these methods frequently require metal catalysts and mediators; most commonly, copper compounds are used in both click chemistry and boronic esters labeling [52,56,72,74]. Copper, along with other commonly used metals such as tin, chromium, and lithium, is classified as a class III element according to ICH guidelines, i.e., considered to have relatively lower toxicity by oral route of administration, but requires risk assessment for parenteral and inhalation routes. While the PDEs of class III elements are relatively high (generally >500 µg/day, specifically 340 µg/day for copper via parenteral route), copper compounds have been frequently used in larger amounts, and therefore copper levels in the final dose must be monitored to ensure compliance. These aspects, which are key elements of the radiochemical production, are only some the points that must be taken into account during the development of a GMP-compliant radiopharmaceutical [106,107,113]. A full quality control system of the final product also needs to be in place, in order to determine radionuclidic, radiochemical, and chemical purity of the dose, in accordance to pharmacopoeias standards. Furthermore, a system of quality assurance is necessary to monitor compliance with standard operating procedures, including batch release approved by a Qualified Person.Taking all these points into account, while a vast selection of PARP-targeted tracers is a constructive contribution to this field, when clinical application is the aim, the intrinsic limitations of the validation process should be kept in mind also during the preclinical development stage.PARP-targeted imaging and radiotherapy is a highly active and rapidly moving field that emerged almost in parallel to the first approvals for PARP inhibitors. Two PARP-targeted PET imaging agents, the olaparib-based [18F]PARPi and the rucaparib-based [18F]FTT, have entered clinical evaluation and more, e.g., [18F]olaparib, are likely to follow. These early clinical studies indicate safety and feasibility of visualizing tumors/quantifying PARP1 expression in a range of tumor types. However, due to the nature of phase 0 and 1 studies, they only include small patient cohorts, and therefore cannot provide conclusive data on their potential to improve clinical standard-of-care yet. Two clinical applications of interest are the selection and monitoring of patients for PARPi therapy (as companion diagnostic) and imaging of tumors that cannot be reliably imaged with existing tracers, such as [18F]FDG. However, additional applications could emerge, including imaging-based response prediction to PARPi or risk stratification of patients. It can be hoped that PARP imaging agents will advance to phase II and III studies based on the encouraging results of early phase studies, to establish their value for specific clinical applications. In addition, we reported on exciting developments in the field of PARP-targeted radiotherapy. Several groups could show promising preclinical anti-tumor efficacy of α- and Auger-emitting agents as monotherapy or in combination with immune checkpoint inhibition. At first look, these agents display moderate tumor uptake compared to considerable uptake in non-target organs (e.g., high PARP1 expressing organs such as the spleen), but toxicity studies showed tolerable safety profiles at efficacious doses in mice. Since PARPi are cell permeable and bind to the PARP1 enzyme once it is bound to damaged DNA in the nucleus, alpha and Auger-emitters decay in close vicinity to the DNA, where they cause lethal damage. The higher PARP1 expression of tumor cell nuclei is a possible explanation for the selective toxicity in tumor cells, but further mechanistic studies are necessary to confirm this or investigate other explanations. Another possible reason is that a certain genetic makeup of tumors is necessary for a high sensitivity to PARP-targeted radiotherapy. Interestingly, while some studies report increased sensitivity in the presence of HR or p53 mutations, other studies report independence of sensitivity from such factors. Overall, more data and improved PARPi-targeted radiopharmaceuticals are eagerly awaited to gain a better understanding of the potential clinical impact of PARP-targeted imaging and therapy. Conceptualization: S.K., N.T.N. and A.P.; writing—original draft preparation: S.K., N.T.N., A.P. and M.N. All authors have read and agreed to the published version of the manuscript.This research received no external funding.SK is a shareholder of Summit Biomedical Imaging, LLC and a co-inventor on PCT application WO2016164771, which describes methods of use of the fluorescent PARP inhibitor PARPi-FL.Olaparib-based PARP-targeted imaging and therapy agents. Modifications from the parent PARPi are highlighted by colored circles. Year of first publication, status of development (in vitro, in vivo) and all related publications are mentioned in their order of appearance in the main text. We also pointed out where clinical imaging studies or preclinical therapeutic results are published.Rucaparib-based PARP-targeted imaging and therapy agents. Modifications from the parent PARPi are highlighted by colored circles. Year of first publication, status of development (in vitro, in vivo), and all related publications are mentioned in their order of appearance in the main text. We also pointed out where clinical imaging studies or preclinical therapeutic results are published.Other PARP-targeted imaging and therapy agents. Modifications from the parent PARPi/parent molecule are highlighted by colored circles. Year of first publication, status of development (in vitro, in vivo) and all related publications are mentioned in their order of appearance in the main text.In vivo evaluation of therapeutic efficacy of PARP-targeted radioligands using therapeutic radioisotopes. See Table 3 for corresponding survival data. (A) Treatment of subcutaneous p53-deficient U87 glioblastoma with 3 × 14.8 MBq intratumoral dose of [131I]PARPi [96]. (B) Therapeutic efficacy of 5 × 74 MBq [123I]MAPi i.v. in subcutaneous p53-deficient HCT116 colorectal cancers compared to [127I]MAPi and vehicle [99]. Stars represent days of treatment. (C) In vivo tumor radiation dosimetry modelling revealed a 150 times higher tumor-nucleus dose per decay of the α-emitter [211At]MM4 compared to the Auger-emitter [125I]KX1 [87]. (D) Treatment of syngeneic glioblastoma (GL26) with a single i.v. dose of ~720 kBq [211At]MM4 and in combination with anti-PD-1 checkpoint inhibitor [100] Stars indicate signifant differences (Anova, *, p < 0.05). Copyright notice: (A) Reprinted with permission from [96]. Copyright 2018, SNMMI. (B) Reprinted with permission [99]. Copyright 2021, American Chemical Society. (C) Reprinted with permission from [87]. Copyright 2020, SNMMI. (D) Reproduced with permission from [100]. Copyright 2021, American Chemical Society.Overview of radioisotopes used in PARP-targeted therapy [86,93,94,95].EC: electron capture; AE: Auger electrons.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01130.txt

@@ -0,0 +1 @@
+SOGUG Multidisciplinary Working Group are listed in acknowledgments.This report presents clinically relevant advances in the management of metastatic bladder cancer, which have been the focus of discussion of expert members of the Spanish Oncology Genitourinary (SOGUG) Multidisciplinary Working Group in the framework of the Genitourinary Alliance project (12GU) designed as a space for the integration of novel information in the care of bladder cancer patients. The present study is focused on different aspects regarding integration of immunotherapy especially in the patient unfit for platinum-based chemotherapy, PD-L1 assays and samples to be evaluated, role of imaging techniques in preoperative staging or re-staging, definition and treatment approach of oligometastatic disease, and rescue strategies in responders. Involvement of a dedicated multidisciplinary team in the care of patients with mBC is crucial to improve outcome.Based on the discussion of current state of research of relevant topics of metastatic bladder cancer (mBC) among a group of experts of a Spanish Oncology Genitourinary (SOGUG) Working Group, a set of recommendations were proposed to overcome the challenges posed by the management of mBC in clinical practice. First-line options in unfit patients for cisplatin are chemotherapy with carboplatin and immunotherapy in PD-L1 positive patients. FDG-PET/CT may be a useful imaging technique in the initial staging or re-staging. In patients with oligometastatic disease, it is important to consider not only the number of metastatic lesions, but also the tumor biology and the clinical course. The combination of stereotactic body radiotherapy and immunotherapy with anti-PD-L1 monoclonal antibodies is under investigation and could improve the results of systemic treatment in patient with oligometastatic disease. Rescue treatment with curative intent could be considered in patients with oligometastatic disease after complete response on FDG-PET/CT. Metastatic disease should be evaluated using the same imaging modality over the course of the disease from diagnosis until rescue treatment. For improving the outcome of patients with mBC, the involvement of a dedicated multidisciplinary team, including urologists, pathologists, oncologists, radiologists and other specialists is of outmost importance in the daily care of these patients.Important advances in the understanding of the molecular mechanisms and tumor progression of urothelial carcinoma have been achieved over the past decade. Management of patients with advanced-stage, unresectable or metastatic bladder cancer (mBC) has shifted in recent years, with novel therapeutic agents available for clinical use, especially new immune checkpoint inhibitors (ICI) directed at programmed cell-death protein 1 (PD-1) or its ligand (PD-L1) with remarkable survival benefits in selected patients with metastatic disease [1]. However, a high unmet need remains for new drugs in platinum-refractory patients with advanced bladder cancer [2].The Genito Urinary Alliance project (I2GU) was designed as a space for the integration of innovation progress in the management of patients with bladder cancer. To this purpose, each expert member of the Spanish Oncology Genitourinary (SOGUG) Multidisciplinary Working Group involved in the project reviewed the literature and redefined the state of art of his/her own area of expertise based on their clinical experience. Controversial and debatable topics of the current knowledge and approach in the care of patients with mBC were also discussed by all expert members of the SOGUG and the topics to be covered by the present review were considered. These include patients unfit for cisplatin-based chemotherapy and integration of immunotherapy, significance and role of PD-L1 assessment, treatment of oligometastatic disease, rescue therapy in respondent patients, and imaging techniques in the evaluation of response. Challenges and recommendations were reached by agreement of all participants to be applicable in clinical practice to facilitate shared decision making for individual patients with metastatic urothelial cancer.The management of bladder cancer requires a multidisciplinary involvement of specialists in urology, medical oncology and radiation oncology to define the appropriate approach for individual patients based on stage and type of cancer. Patients with mBC account for 5% of newly diagnosed cases [3]. Platinum-based chemotherapy (cisplatin, carboplatin) has been for decades the treatment of choice in mBC. Approximately 40% of patients with adequate renal function are eligible for chemotherapy with cisplatin, 40% unfit for cisplatin are eligible for chemotherapy with carboplatin, and the remaining 20% are unfit for any platinum-based chemotherapy and may be treated with different options: monotherapy with paclitaxel, gemcitabine, and others [4,5,6]. In an effort to develop a consensus definition of patients with mBC unfit for cisplatin-based chemotherapy, a working group was assembled and conducted a survey of 120 international academic and community-based genitourinary oncologists [7]. Proposed eligibility criteria for mBC patients unfit for cisplatin-based chemotherapy include at least one of the following: WHO or ECOG performance status of 2 or Karnofsky status of 60–70%, creatinine clearance < 60 mL/min, Common Terminology Criteria for Adverse Events (CTCAE) v4 grade ≥ 2 audiometric hearing loss or ≥2 peripheral neuropathy, and New York Heart Association (NYHA) class III heart failure. Age is not included among the definition criteria but is an important factor to be considered in daily practice (e.g., elderly patients with comorbidities).Renal dysfunction, poor performance status, and comorbidities may preclude frontline cisplatin-based chemotherapy in clinical practice. In the experience of a community-based cancer center of 298 patients with mBC, a first-line cisplatin-based regimen was administered to 35.9% of patients, carboplatin-based to 27.2%, non-platinum-based chemotherapy to 8.4%, and no chemotherapy in 23.8% [8]. In the IMvigor130 phase III clinical trial [9] carried out in untreated patients with mBC and randomized to atezolizumab plus platinum-based/gemcitabine chemotherapy, atezolizumab monotherapy, or placebo plus platinum-based/gemcitabine chemotherapy, 45%, 30% and 35% of patients in each group were ineligible for cisplatin-based chemotherapy, but 70%, 63% and 66% received chemotherapy with carboplatin instead of cisplatin, which may reflect real-world clinical practice.In patients eligible for cisplatin-based chemotherapy, there are several combinations for first-line treatment, with gemcitabine/cisplatin (GC) as the most common. In a large randomized phase III study of GC versus methotrexate/vinblastine/doxorubicin/cisplatin (MVAC) [10], GC provided similar efficacy in terms of overall survival and progression-free survival compared with MVAC, but with a superior safety profile. Gemcitabine/carboplatin is also the most frequently used combination in patients unfit for cisplatin-based chemotherapy [11]. In second-line treatment, vinflunine showed a marginal efficacy as compared with best supportive care in a phase 3 clinical trial [12] and was not approved by the Food and Drug Administration (FDA) but received approval of the European Medicines Agency (EMEA). Standard therapy in mBC before the introduction of immunotherapy showed response rates of 40–50% and median survival of 12–15 months for first-line chemotherapy (GC, MVAC and paclitaxel/cisplatin/gemcitabine) in cisplatin eligible patients, 36–56% and 7–9 months for gemcitabine/carboplatin in cisplatin ineligible patients, and about 10% and 5–8 months for the single agent vinflunine in second-line therapy.Following the requirements of the EMEA, in mBC patients ineligible for cisplatin-containing chemotherapy, PD-1/PD-L1 therapy with either atezolizumab or pembrolizumab requires the use of an FDA-approved companion diagnostic test to determine PD-L1 levels in tumor tissue. In patients with locally advanced and unresectable or mBC, two single-arm multicenter phase II studies evaluated the use of anti-PD-L1 therapy as first-line therapy in cisplatin-ineligible patients. In the IMvigor210 clinical trial of atezolizumab, at a median follow-up of 17.2 months, the objective response rate was 23% and the median overall survival was 15.9 months [13], whereas in the KEYNOTE-052 trial of pembrolizumab up to a median follow-up of 5 years, the objective response rate was 28.9% [14]. Also, in both studies durable responses were obtained. However, despite these encouraging results, response rates, progression-free survival and overall survival associated with ICIs have not been proven to be superior to carboplatin-based chemotherapy, and carboplatin-based chemotherapy remains a viable first-line treatment option in cisplatin-ineligible PD-L1-positive patients with mBC until mature data from randomized phase III of ICIs will become available [15]. Main results from first-line phase II and III trials of anti-PD-L1 agents in advanced urothelial cancer [13,16,17,18,19,20] are listed in Table 1.In a study of maintenance treatment with the anti-PD-L1 antibody avelumab in patients who did not have disease progression with first-line chemotherapy (four to six cycles of GC or gemcitabine/carboplatin), avelumab was associated with statistically significant improvements in overall survival at 1 year as compared with best supportive care in the whole study population (hazard ratio (HR) 0.69, 95% confidence interval (CI) 0.56–0.86) and in the PD-L1-positive population (HR 0.56, 95% CI 0.40–0.79) [19].Avelumab, durvalumab, nivolumab, atezolizumab and pembrolizumab are FDA-approved ICIs that have been evaluated as second-line options, but only atezolizumab [21] and pembrolizumab [22] in the framework of phase III randomized studies. Median overall survival was around 11 months, 20% response rate, and durable response rates at 2 years of approximately 40%.In relation to targeted therapy, erdafitinib, an oral pan-fibroblast growth factor receptor (FGFR)-targeted agent based on relevant clinical activity in mBC patients whose tumors bear actionable FGFR alterations, data of an open-label phase 2 study in 99 patients showed confirmed response in 40% (complete response 3%, partial response 37%) [23]. The median duration of response was 5.6 months and approximately 30% of these responses were maintained for more than 12 months. At 12 months, the rate of overall survival was 55% and the rate of progression-free survival was 19%.In the open-label phase 3 study of enfortumab vendotin (EV), an antibody-drug conjugate targeting nectin-4 was administered to patients with locally advanced or mBC with disease progression during or after treatment with PD-1/PD-L1 inhibitors and compared to chemotherapy [24]. The primary endpoint was overall survival. The median overall survival was 12.88 in the EV group vs. 8.97 months in the chemotherapy group, with a hazard ratio for death of 0.70 (95% CI 0.56–0.89; p = 0.001). EV significantly prolonged survival as compared with chemotherapy.A summary of the role of immunotherapy in mBC is shown in Table 2.To improve definition of “unfit” considering difficulties in the integral assessment of older patients and adequate initial evaluation of the patient’s general status.Consolidated data are needed to determine the superiority of immunotherapy over chemotherapy for first-line treatment.PD-L1 expression should be measured for the selection of first-line treatment in “unfit” patients.In the second-line setting, immunotherapy is preferred to chemotherapy, independently of the status of PD-L1.The use of molecular classification based on gene expression profiles to guide therapeutic decisions in clinical practice is still limited and homogenized terminology is needed.Assessment of PD-L1 levels in tumor tissue is currently recommended for a better selection of candidates for first-line treatment with the anti-PD-L1 agents atezolizumab or pembrolizumab in patients with locally advanced urothelial cancer, mBC or no candidates/refractory to cisplatin-based chemotherapy. Indication of PD-L1 testing was established based on the “one test, one drug” model, with specific quantification and interpretation criteria, and the companion diagnostic tests associated with the clinical response to the anti-PD-L1 drug. Table 3 summarizes assays for PD-L1 expression in urothelial cancer [25].Before an PD-L1 assay, it is important to define the information that should necessary to include in the testing request form for practical, technical and cost-efficiency reasons (e.g., specification of the test, type of anti-PD-L1 drug that is intended to be prescribed, type of sample, controls). In relation to the selection of the most appropriate sample (Table 4), it is recommended the use of the most recent specimen with sufficient tumor tissue and a lower level of cauterization and necrosis. In selected cases, the use of various blocks (and even various fields) may be necessary to assess heterogeneity. There is no uniform agreement regarding the use of specimens from the primary tumor or metastatic sites, but the use of samples after neoadjuvant chemotherapy is discouraged. Positive (tonsils) and negative controls are recommended and should always be carried out according to the manufacturer’s instructions.These assays, however, use different antibodies, immunohistochemical protocols, scoring algorithms, and cutoffs to define high/low PD-L1 expression in urothelial cancer, so that it is necessary to determine whether different therapeutic decisions may be related to the use of specific antibodies may involve different therapeutic decisions. Different studies have shown that SP263, 22C3 and 28-8 assays are analytically similar with high correlation coefficients [29,30,31,32], in contrast to the SP142 assay that shows divergent staining results, pooled percentage of agreement of 59% with SP263, 22C3 and 28-8, and fewer eligible patients identified for first-line therapy with atezolizumab [33]. Thus, patient selection for UC-1 L treatment with pembrolizumab or atezolizumab requires the use of an FDA/CE-IVD approved assay. The SP142 and 22C3 clones were approved in assays for this purpose.The concordance of the four PD-L1 expression assays had been also evaluated in primary and metastatic bladder carcinomas. Two studies of matched pairs of transurethral resections of the bladder (TURB), cystectomy specimens and lymph node metastases showed concordant overall results, with discordance occurring more frequently after neoadjuvant therapy [34,35]. In relation to the characterization of immunohistochemical markers to recognize basal and luminal molecular subtypes, it has been shown that the basal subtype high-grade urothelial cancer has abundance of CD8+ T cells with increased expression of inhibitory markers [36]. In the pure urothelial carcinoma histology, which accounts for up to one-third of advanced cases, the three SP263, 22C3 and SP142 clones showed strong agreement in pairwise comparisons of tumor and immune cells, with high expression in urothelial carcinomas with squamous differentiation and lymphoepithelioma-like variants [37,38,39].Training of pathologists is indispensable to reduce intra- and inter-observer variability in PD-L1 expression assays.The use of specimens after neoadjuvant chemotherapy should be avoided and in case of high staining heterogeneity, examination of different fields and blocks is recommended.The pathological report should include integrated histological and immunohistochemical information with technical details (e.g., antibodies, platforms, cutoffs definitions) and quantitative values of the percentages of PD-L1 positivity, with a final recommendation regarding eligibility for anti-PD-L1 treatment.Large validation studies preferably including patients treated with ICIs are necessary to increase the use of PD-L1 assays in clinical practice.Assessment of molecular subtypes in addition to PD-L1 expression as well as tumor mutational load, CD8+ T cells, M2 macrophages, CTL4, and TNFβ in tumor-related fibroblasts are also promising areas for future studies.The main questions regarding the role of imaging techniques in patients with mBC refer to the indications of 18F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) in preoperative staging or re-staging (recurrence) in patients with suspicion of metastases, as well as in the assessment of response in patients with metastatic disease.The main advantage of FDG-PET/CT as compared with morphological images is the detection of bone metastatic disease. Although lytic bone lesions can be detected by CT, other lesions without density alterations can be evaluated by FDG uptake (Figure 1).The usefulness of FDG-PET/CT in small lesions of <1 cm (especially in organs in motion such as the lungs) is limited, with TC or magnetic resonance imaging (MRI) providing a higher resolution, although all imaging techniques have limitations in the assessment of lesions of less than 5 mm.The presence of lymph node involvement and distant metastasis in patients with invasive bladder carcinoma is a major determinant of survival and, therefore, a pivotal element in the therapeutic management. The rate of disseminated disease is very high increasing from 25% in T2 stage to 50% in T3 stage. Also, 50% of patients with local disease undergoing radical cystectomy will develop distant metastasis at 2 years [40]. Despite radical treatment, the 5-year overall survival of 50% probably indicates that spread of tumor cells had occurred before surgery [41].The use of FDG-PET/CT in staging of the primary tumor currently lacks sufficient evidence for recommendation. However, the EUA-ESMO consensus statements on the management of advanced and variant bladder cancer, recommends the use of FDG-PET/CT scanning in oligometastatic disease staging when considering radical treatment [15]. FDG-PET/CT is also recommended in patients with lymph node involvement outside the pelvis or indeterminate/suspected metastatic lesions in high-risk patients [42]. In the 2020 NCCN guidelines on bladder cancer, bone scan, MRI and FDG-PET/CT is recommended to evaluate the extent of disease in symptomatic patients, at high risk of metastases or positive biomarkers of bone disease [43]. In a study of patients with advanced disease, FDG-PET/CT showed a sensitivity of 87% and specificity of 88% for the organ-based analysis and 81% and 94% for the patient-based analysis [44]. Moreover, pre- and post-PET surveys revealed that FDG-PET/CT detected more malignant disease than conventional CT/MRI in 40% of patients, and post-PET surveys showed that clinicians changed their planned management in 68% of patients based on the FDG-PET/CT results [44]. However, prospective comparative studies assessing the diagnostic reliability of the different imaging techniques for staging in advanced bladder cancer are lacking.CT is the standard technique in the assessment of response in patients with metastatic disease, and although Response Evaluation Criteria in Solid Tumors (RECIST) has been the most widely accepted method for assessing tumor response, measurement of unidimensional diameters is a limitation of RECIST. Other limitations include changes in tumor form, assessment of unmeasurable lesions such as bone lesions, cystic transformation, heterogeneous response or no volume changes, and inability to detect tumor burden. The European Organisation for Research and Treatment of Cancer (EORTC) criteria have proven to be more sensitive in detecting complete and partial remission when compared to RECIST criteria, but hypermetabolic lesions other than tumor itself, physiological urinary FDG activity, or evaluation of organs that have high glucose utilization are limitations of EORTC [45]. On the other hand, studies assessing metabolic tumor burden are lacking.FDG-PET/CT is recommended in the initial staging or re-staging, in oligometastatic disease staging when considering radical treatment. FDG-PET/CT is also recommended in patients with lymph node involvement outside the pelvis or in case of indeterminate/suspected metastatic lesions in high risk patients.In centers in which FDG-PET/CT is not available, morphological images are also an option.Imaging techniques (CT, MRI, FDG-PET/CT) are complementary and it is important to select the appropriate imaging method in each case.The imaging technique used in the treatment follow-up of patients should be the same to that initially performed in the assessment previous to treatment.Studies of FDG-PET/CT using techniques for assessing metabolic tumor burden (metabolic tumor volume, glycolysis total rate) are recommended.In 1995, Hellmann and Weichselbaum [46] proposed a clinical state of metastasis termed “oligometastases” that refers to restricted tumor metastatic capacity. According to this concept, there is an intermediate biological state of restricted metastatic capacity in which spread may be limited to specific organs and metastases may be present in limited numbers. This transitional state to dissemination may have the clinical implication that some patients affected of a significant oligometastatic state should be amenable to curative therapeutic strategies [47]. The original tumor may be controlled or uncontrolled. Based on the concept of oligometastases, the proposal of “oligo-recurrence” has a similar notion and includes the conditions of a primary site of the cancer controlled, one to several distant metastases/recurrences (usually one) in one to several organs (usually one), and one to several distant metastases/recurrences can be treated with local therapy [48]. The concept of oligo-progressive disease defines recurrence or limited progression after cytoreductive therapy or following systemic treatment, and there is relapse in a limited number of regions.A group of international experts in diagnosis and treatment of oligometastatic disease from the EORTC and European Society for Radiotherapy and Oncology (ESTRO) OligoCare project participated in a consensus process on characterization and classification of oligometastatic disease and established a system nomenclature that cover all possible clinical situations of imaging findings with few metastases [49]. This classification based on decision tree analysis includes the three main categories of “de novo”, “repeat” and “induced” oligometastatic disease. De novo oligometastatic disease differentiates synchronous and metachronous oligo-recurrence, repeat oligometastatic disease involves response to local treatment and a small number of recurrences after a treatment-free interval, and induced oligo-progression is characterized by good responses to systemic therapy of polymetastatic disease but only a few metastases develop resistance and progress later on.However, the optimal imaging modalities for the diagnosis and response prediction in patients with oligometastatic disease remain to be determined. Prognostic factors identified in the setting of oligometastatic disease in patients with mBC after total cystectomy include M1 disease with node-only involvement and good performance status as compared to visceral metastases (bone, lung, liver) and/or poor performance status, a solitary metastatic organ, number of metastatic lesions 3 or less, the largest diameter of metastatic foci of 5 cm or less, and no liver metastasis [50].In relation to management of patients with oligometastatic disease, outcome is more favorable for treated than untreated patients, although no level 1 evidence is available and almost all studies are retrospective aimed to consolidate response to a previous systemic treatment or, in some cases, delay in the beginning of systemic therapy, with FDG-PET/CT as the imaging technique most used [51,52]. In a systematic review and meta-analysis to explore the role of metastasectomy in mBC based on data from 17 studies and 412 patients, metastasectomy displayed a significant better overall survival in comparison to non-surgical treatment of metastatic lesions, but only five studies were included in the meta-analysis [53]. Also, reporting of systemic treatment type, treatment schedules, and response to treatment were heterogeneous, and all except for three studies, were retrospective and non-randomized leading to a high risk of bias.Stereotactic body radiotherapy (SBRT) associated with immunotherapy using checkpoint inhibitors (atezolizumab, avelumab, durvalumab, nivolumab, pembrolizumab) can enhanced the abscopal effect due to SBRT, and could improve the results of systemic treatment. In a randomized phase 1 trial combining pembrolizumab with either sequential or concomitant SBRT in mBC, no dose-limiting toxicity occurred and an overall response rate of 44.4% in concomitant SBRT was observed [54]. However, predicting a response is intricate with no single marker (PD-L1 or tumor burden) being sufficient to explain response or survival. Multifactorial approach combining tumor-specific and immune markers might be the key to identify who will benefit from treatment, with circulating tumor DNA (ctDNA) fraction that may serve as a surrogate for monitoring disease evolution.It is important to understand the concept of oligometastatic disease and to consider not only the number of metastatic lesions, but also the tumor biology and the patient’s clinical course.The use of ablative radiation therapy followed by systemic treatment is a recommendable strategy for the treatment of oligometastatic bladder cancer.Patients with M1 disease including node involvement only and good performance status have a better prognosis than patients with either visceral metastasis and/or poor perforformance status.Further advances in the combination of SBRT and immunotherapy with anti-PD-L1 monoclonal antibodies could improve the results of systemic treatment in patient with oligometastatic disease.In general, rescue treatment is considered in patients with metastatic disease or clinically positive lymph nodes (with locally advanced disease) who had presented a clear response to systematic treatment. Rescue treatment is considered in patients with locally advanced disease or metastatic disease (synchronic or oligometastatic), which should be differentiated from recurrence appearing during the course of the disease after a previous radical treatment. Patients with recurrence are candidates to surgical rescue without systemic treatment (in case of oligometastases or metachronic metastatic disease).There is little information in the literature regarding the role of surgery in removing metastatic lesions after response to chemotherapy. In a systematic review of 28 selected articles [55], surgery in patients with clinically positive lymph node based on data from 11 studies was associated with pooled percentages of 33%, 44% and 18% for complete clinical response, partial clinical response and pathological response, respectively. A few studies evaluated metastasectomy in lung metastasis, retroperitoneal lymph nodes and other metastatic sites, as well as cytoreductive radical cystectomy. There are important differences among studies in the percentages of partial (5–57%) and complete (24–35%) clinical response, inconsistencies in reporting pathological response, pT0 (9–30%), pN0 (37–55%), no significant differences in overall survival between cN1 and cN2-3, and with a median follow-up of 13–60 months, 5-year cancer-specific survival varies between 23% and 63%. However, as a result of a number of limitations, such as retrospective reviews of single or multiple-institution data sets, small series and span many years, unclear definition of patients who should receive cytoreductive surgery, variations in the extent of surgery, and outcomes reported for disease involvement from various regions, there are no definitive indications as to when and to who apply postchemotherapy surgery [56]. It remains challenging to provide a definitive estimate of the magnitude of this benefit from the literature, as this will vary according to the site (s) of disease and the initial tumor burden. According to NCCN guidelines [43], consolidation cystectomy or consolidation chemoradiotherapy should be offered in selected patients with complete response.Before planning surgical rescue treatment, methods for assessing clinical response is heterogeneous, with pathological complete response (pCR) used in clinical trials based on neoadjuvant therapy with ICIs [13,57,58,59] and objective response rate (ORR) in clinical trials focused on treatment of M1 patients [60,61,62] (Figure 2).Therefore, the type of response should be defined considering different factors, such as the percentage of reduction of the initial tumor, the percentage of response of the primary tumor versus metastasis, partial response versus complete response, and type of imaging technique for assessment. In a Delphi survey study under the auspices of the EUA-ESMO Guidelines Committees [15], statements related to the role of treatment with curative intent in oligometastatic disease were discussed, with the following three proposed statements achieving consensus: (1) in a minority of patients with one metastatic lesion, cure is possible after radical treatment, (2) PET-CT scanning should be included in staging when considering radical treatment, and (3) radical treatment should be accompanied by adjuvant or neoadjuvant systemic therapy. Moreover, other statements referred to the fact that liver and bone are unfavorable oligometastatic sites for curative therapy, cure is not possible in the presence of two metastatic sites, and in metachronous oligometastatic disease, time to relapse in an important prognostic indicator.The benefit of consolidation surgery in overall survival and cancer-specific survival in metastatic bladder disease remains to be defined.Rescue treatment with curative intent could be considered in patients with oligometastatic disease after complete response on FDG-PET/CT.Metastatic disease should be evaluated using the same imaging modality over the course of the disease from diagnosis until rescue treatment.Rescue treatment includes radical cystectomy and lymphadenectomy or consolidation surgery based on radical cystectomy, lymphadenectomy and metastasectomy.It is necessary to establish a precise prediction of response as well as to define partial response considering volume, site, number of cycles and biomarkers.A tight definition of a patient unfit for cisplatin treatment remains a challenge in clinical practice. The patient’s age and different comorbidities are additional factors to be considered. In cisplatin-ineligible patients, both chemotherapy with carboplatin-gemcitabine and ICIs (atezolizumab or pembrolizumab) are valid first-line options. However, there is still no evidence of the superiority of ICIs as compared with chemotherapy.The role of PD-L1 as a predictive biomarker of response in urothelial cancer remains unclear, but its determination is mandatory for the selection of candidates to be treated with ICIs in cisplatin-unfit patients. Training of pathologists is indispensable to reduce intra- and interobserver variations in PD-L1 expression assays. Analysis of the most recent sample with the most representative tumor material available is recommended. Although there is no agreement regarding the use of specimens from the primary tumor or metastatic sites, using samples after neoadjuvant chemotherapy is discouraged.FDG-PET/CT is recommended in the initial staging or re-staging as well as in oligometastatic disease staging when radical treatment is considered. FDG-PET/CT is also recommended in the presence of lymph node involvement outside the pelvis or indeterminate/suspected metastatic lesions in high-risk patients. The usefulness of FDG-PET/CT in small lesions of <1 cm (especially in organs in motion such as the lungs) is limited, with TC or MRI providing a higher resolution.In oligometastatic disease, it is important to define correctly not only the number of metastases, but also the biology and clinical course of the disease. The use of ablative radiation therapy followed by systemic treatment is a recommendable strategy for the treatment of oligometastatic bladder cancer. The combination of immunotherapy and SBRT is an investigational strategy with very promising results.The benefit of consolidation surgery in overall survival and cancer-specific survival in metastatic bladder disease remains to be defined. Rescue treatment with curative intent could be considered in patients with oligometastatic disease after complete response on FDG-PET/CT.Regarding the role of the hospital pharmacist, training of healthcare workers and diffusion of information are key factors for improving safety in the administration of treatment in bladder cancer patients. Chemotherapeutic agents and immunotherapy drugs are considered hazardous drugs. Appropriate measurements should be taken to minimize exposure of healthcare professionals during drug administration.Finally, for improving the outcome of patients with mBC, the involvement of a dedicated multidisciplinary team, including urologists, pathologists, oncologists, radiologists and other specialists is of outmost importance in the daily care of these patients.Conceptualization, all authors; Definition of the Patient Unfit and Integration of Immunotherapy in the Management of Advanced Urothelial Cancer, A.G.-d.-A.; PD-L1 Testing in Urothelial Carcinoma, P.G.-P.; Role of Imaging Techniques in Metastatic Urothelial Cancer, A.M.G.V.; Treatment of Oligometastatic Disease, A.J.C.-M.; Systematization of Rescue Treatment Strategies in Responders, E.L.-E.; review and validation, M.Á.C.; writing and review, all authors. All authors have read and agreed to the published version of the manuscript.This research was funded by AstraZeneca.The authors thank Marta Pulido, for editing the manuscript and editorial assistance, and BCNscience for coordinating the research. Membership of the Spanish Oncology Genitourinary (SOGUG) Working Group: Aránzazu González del Alba, Antonio José Conde-Moreno, Ana M. García Vicente, Pilar González-Peramato, Estefanía Linares-Espinós, Ferran Algaba, José Ángel Arranz Marga Garrido, Regina Gironés Sarrió, Enrique Gallardo, Antonio Gómez Caamaño, Eva González-Haba, Antonio López-Beltran, Paula Pelechano, José Antonio Marcos-Rodríguez, Pablo Maroto, Alfredo Rodríguez Antolín, José Rubio-Briones, Julián Sanz, María Almudena Vera González, Almudena Zapatero and Miguel Ángel Climent.A. González-del-Alba: has received research funding from Astellas, travel grants from Astellas, Jansen, Sanofi, BMS, Roche, Pfizer and Ipsen and honoraria for speaker engagements, advisory boards and continuous medical education from Astellas, Janssen, Sanofi, Bayer, Roche, Ipsen, BMS, MSD, Pfizer, Eusa Pharma, Eisai and Astra Zeneca; A.J. Conde-Moreno: declares no conflicts of interest; A.M. García Vicente: declares no conflicts of interest; P. Gónzález-Peramato: declares no conflict of interest; E. Linares-Espinós: has participated in a Trial with Merck; M.A. Climent: consulting role for BMS, MSD, Bayer, EUNSA, Pfizer, Roche, Janssen, Pierre Fabre, Ipsen; honoraria from BMS, Astellas, Janssen, MSD, Sanofi, Bayer, Roche, Pfizer, Novartis, Ipsen; travel/accommodation expenses from Janssen, Astellas, Roche, Ipsen, MSD.FDG-PET/CT showed metastatic bone lesions (a) undetected by CT (b).Differences in pCR in clinical trials with neoadjuvant ICIs in M0 patients and ORR in patients with M1 disease treated with ICSs alone or combined, data from [13,52,53,54,55,56,57].First-Line Phase II and III trials of anti-PD-L1 agents in advanced urothelial cancer.1 L = first-line; Atezo = atezolizumab; Avelu = avelumab; BSC = best supportive care; ChT = chemotherapy; CI = confidence interval; Cis = cisplatin; D = durvalumab; HR = hazard ratio; ITT = intention-to-treat; NE = not estimable; Pembro = pembrolizumab; OS = overall survival; Pbo = placebo; PFS = progression-free survival; T = tremelimumab.Summary of the role of anti-PD-L1 agents in advanced urothelial cancer.New FDA/EMA approvals:Pembrolizumab in first-line PD-L1 + cisplatin-ineligible and second-line mBC.Atezolizumab in first-line PD-L1 + cisplatin-ineligible and second-line mBC.Pembrolizumab is the first agent to ever show overall survival benefit in second-line therapy for mBC.Maintenance treatment with Avelumab (anti-PD-L1) in patients who have not progressed to first-line platinum achieves improvement in overall survival.PD-L1 expression does not guide treatment selection in second-line treatment in urothelial cancer, because benefit with treatment has been shown in the overall population.PD-L1 expression would “a priori” increase the likelihood of benefit from ICIs in different clinical settings, although results of studies are conflicting to validate PD-L1 as a predictor of response.Robust predictive biomarker is still lacking.Currently available assays for PD-L1 expression testing before treatment with anti-PD-L1 agents in patients with locally advanced or metastatic urothelial cancer.IHC: immunohistochemistry; IC: immune cells; TC: tumor cells; ICS: immune cells score; CPS: combined positive score; bibliographic references in brackets.Samples to be evaluated for PD-L1 expression testing.TRU: transurethral resection; UC: urothelial cancer; CIS: carcinoma in situ.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01131.txt

@@ -0,0 +1 @@
+Phase I clinical trials are a cornerstone of pharmaceutical development in oncology. Many studies have now attempted to incorporate pharmacogenomics into phase I studies; however, many of these studies have fundamental flaws that that preclude interpretation and application of their findings. Study populations are often small and heterogeneous with multiple disease states, multiple dose levels, and prior therapies. Genetic testing typically includes few variants in candidate genes that do no encapsulate the full range of phenotypic variability in protein function. Moreover, a plurality of these studies do not present scientifically robust clinical or preclinical justification for undertaking pharmacogenomics studies. A significant amount of progress in understanding pharmacogenomic variability has occurred since pharmacogenomics approaches first began appearing in the literature. This progress can be immediately leveraged for the vast majority of Phase I studies. The purpose of this review is to summarize the current literature pertaining to Phase I incorporation of pharmacogenomics studies, analyze potential flaws in study design, and suggest approaches that can improve design of future scientific efforts.While over ten-thousand phase I studies are published in oncology, fewer than 1% of these studies stratify patients based on genetic variants that influence pharmacology. Pharmacogenetics-based patient stratification can improve the success of clinical trials by identifying responsive patients who have less potential to develop toxicity; however, the scientific limits imposed by phase I study designs reduce the potential for these studies to make conclusions. We compiled all phase I studies in oncology with pharmacogenetics endpoints (n = 84), evaluating toxicity (n = 42), response or PFS (n = 32), and pharmacokinetics (n = 40). Most of these studies focus on a limited number of agent classes: Topoisomerase inhibitors, antimetabolites, and anti-angiogenesis agents. Eight genotype-directed phase I studies were identified. Phase I studies consist of homogeneous populations with a variety of comorbidities, prior therapies, racial backgrounds, and other factors that confound statistical analysis of pharmacogenetics. Taken together, phase I studies analyzed herein treated small numbers of patients (median, 95% CI  = 28, 24–31), evaluated few variants that are known to change phenotype, and provided little justification of pharmacogenetics hypotheses. Future studies should account for these factors during study design to optimize the success of phase I studies and to answer important scientific questions.For approximately 20 years, pharmacogenomics approaches have been appearing in phase I clinical trials of anticancer medications. Accounting for genetic variability in early clinical development is worthwhile for agents in which marker-based patient selection is likely to improve success by identifying responsive and lower-risk populations [1]. This is particularly true for oncology agents, which have the highest attrition rates in clinical development and are the most likely to benefit from patient stratification [2]. Yet, the scientific constraints imposed by phase I study designs also limit the usefulness of such approaches [3]. How can reproducible or generalizable results be generated in small, heterogeneous, heavily pretreated populations that are administered combinations of various medications? Can these limitations be overcome to produce robust clinical analyses accounting for genetic variation in dose optimization? Constructive criticism of published phase I trials incorporating pharmacogenomics is warranted, and many lessons can be learned by examining the performance of such studies over two decades.Following drug discovery, lead optimization is conducted in a limited set of molecules that undergoes testing for efficacy, pharmacokinetics (PK), and toxicity in model systems. Lead compounds are screened based on desirable properties associated with potential clinical utilization [4]. Such studies utilize information gathered at the bench to apply a given therapeutic to an appropriate cohort of patients in the clinical setting, and they are becoming increasingly precise. For example, traditional cancer cell lines are now being scrutinized for their applicability to human cancer in situ, which has resulted in improvements in the prioritization of therapeutic targets and drug molecules based on several genomic considerations [5,6,7].Characterization of the absorption, distribution, metabolism, elimination, and activation (ADME-A) properties of compounds is also exceedingly important in preclinical characterizations of drug candidates since both the ability of a bioactive drug to reach the intended target and its toxicity depend on pharmacokinetic properties [4]. In vitro, in vivo, and in silico ADME-A screening techniques have become increasingly sophisticated, and many of these methods provide precise information about genetic variables that are associated with drug disposition [8,9]. In many cases, reverse translation of prior clinical experience can also be included in preclinical models that clarify the mechanistic basis of clinical observations [10].Following discovery and preclinical characterization, molecules that are still suitable for human use move to the phases of drug development, including clinical testing [4]. A typical phase I study design involves escalating a dose that was previously determined in animal testing. The decision to increase or decrease dose is based on the presence or absence of severe toxicity at each dose level. This approach does not require assumptions about the dose-toxicity curve; however, it may expose certain populations to greater risk of toxicity should prior knowledge about variants that affect drug pharmacokinetics (PK) or pharmacodynamics (PD) be available [11]. Oftentimes, such knowledge is available from preclinical models or, perhaps more often, from retranslating prospective or retrospective analysis of clinical trial data. When decision-making is focused on target variability, patient specific factors, and PK/PD modeling, significant improvements in Phase III completion are observed [1]. These strategies include patient stratification early in the drug development process and marker-based patient selection [1,12]. Thus, appropriate application of knowledge in early clinical development reduces negative impacts on patients while simultaneously improving the attrition rate of medications undergoing development.Despite the narrow therapeutic index of anticancer agents and the frequent need to administer these medications at high dose to avoid inefficacy, pharmacogenetic approaches are rare in the early development of oncology agents. Sufficiently powered studies with adequate genetic coverage in appropriate populations are even rarer. Why do so few studies incorporate pharmacogenetics approaches in Phase I designs, and why do so many of these studies fail to detect an association? [3] The purpose of this review is to provide an overview of currently published phase I studies incorporating germline pharmacogenomics approaches and explore the potential for improving pharmacogenomics strategies in future phase I studies.Using “Clinical Trial, Phase I” filter in https://pubmed.ncbi.nlm.nih.gov, we searched for the following terms: “pharmacogenetics cancer”, “pharmacogenomics cancer”, “polymorphism cancer”, “pharmacogenetics leukemia”, “pharmacogenomics leukemia”, “pharmacogenetics oncology”, “pharmacogenomics oncology”, and “polymorphism oncology”. The final search for these studies was conducted on 21 January 2022. Studies were included if they contained data about at least one commonly inherited germline genetic variant. Studies were excluded if they only pertained to cancer mutations (i.e., companion diagnostics) and/or gene expression. Of 11,737 phase I clinical trials published on the subject of “cancer”, and 14,247 phase I clinical trials mentioning “oncology”, we found only 84 different phase I, phase Ib, and phase I/II clinical trials that met the above criteria (0.72% and 0.59% of studies, respectively). All studies utilized the candidate gene approach, and no study included hypothesis-free methods. The present analysis includes studies regardless of prospective or retrospective design provided a gene–drug pair was tested in a cohort of patients participating in phase I clinical testing of an anticancer agent. Characteristics of the studies are presented in Table 1.More phase I studies we examined have compared genetic variants to drug toxicity than any other endpoint (n = 116 comparisons in 42 studies), and every one of these studies evaluated genes involved in the ADME-A or activity pathway of drugs under study (Figure 1). For example, the most frequent genes studied versus toxicity include UDP-glucuronosyltransferases (UGTs) that conjugate glucuronides to a variety of medications (n = 21 comparisons with genotype) and ATP-binding cassette transporters (ABCs) that convey several drug types across biological membranes (n = 15 comparisons; Table 2). As expected, fewer studies have evaluated pathways that are related to specific classes of drugs, such as the relationship between variants in Aurora Kinase A and B (AURKA and AURKB) and the AURK inhibitor, danusertib (n = 1 study). Studies of genetics versus response or PFS are rarer (n = 73 comparisons in 31 studies), but they also pertain to a mixture of genes involved in both pharmacokinetics and pharmacodynamics.Thirteen of the 42 pharmacogenetics studies involving toxicity did not conduct a formal statistical analysis, and 11 of 32 studies related to response or PFS pharmacogenetics did not analyze data they collected (Table 2). Of the remaining 29 pharmacogenetics studies evaluating toxicity, only seven studies found an association with toxicity (18 comparisons) and 22 studies found no association (72 total comparisons). In general, low coverage was observed within each gene (median = 1; range 1–5) in few patients (median 24.5; range 10–111) at multiple dose or treatment levels (median 3 dose or treatment levels; range 1–13 levels). Of those studies analyzing response or PFS, nine of 21 studies detected an association with a genetic variant (11 comparisons) and 12 did not (47 comparisons). A median of 1 variant was detected in each gene (range 1–30) in a median of 30 patients (range 10–115) at a median of 3 dose or treatment levels (range 1–12).Of those studies that have evaluated genetic variants in ADME-A genes or genes involved in drug action (Table 2), a median of two variants were probed per gene (range 1–61 variants). Only three studies evaluated more than 10 variants in genes involved in Phase I or II metabolism [13,14,15]. Yet, moderate to definitive evidence exists for at least 16 star alleles in CYP2A6, seven in CYP2C19, 20 in CYP2C9, 26 in CYP2D6, six in CYP3A4, three in CYP3A5, 16 in NAT2, and five in UGT1A1 according to pharmgkb.org. Moreover, the genotype-predicted activity status (e.g., ultrarapid, rapid, extensive, intermediate, or poor metabolizer) of most of these genes is now available, but this information is not being used routinely in phase I studies (Table 2).Twenty of the 40 studies that compared genotype to pharmacokinetics never conducted a formal statistical analysis (data for one study were not disclosed), instead offering an observational commentary about specific patients harboring certain genetic variants (Table 2). Of the remaining 20 studies, 13 (61.9%) found a relationship between a genetic variant and the pharmacokinetic properties of a medication (15 comparisons with genotype) and seven studies did not (43 comparisons). Of these, five studies pertained to the relationship between irinotecan (or SN38) and UGT1A1 variants, a gene–drug interaction that is well characterized in the scientific literature with multiple iterations of retranslation [16]. A median of 28 patients were included in these studies (range 10–94) at a median of three different doses or treatments (range 1–12).Studies examining the statistical relationship between pharmacokinetics and genotype demonstrate a higher ratio of statistical associations per comparison (14/59 comparisons with genotype, 23.7%) than those focused on toxicity (18/93 comparisons, 19.4%) or response/PFS (11/67 comparisons, 16.4%; Table 2), although the difference in these ratios was not statistically significant (p = 0.59, chi-squared test). If all endpoints are considered together, a statistically significant relationship is apparent between a higher number of patients tested and detection of an association with genotype (median = 28 patients in non-associations, median = 34 patients in associations; p = 0.020; Wilcoxon rank sum test). Statistical positivity in toxicity studies was also associated with the number of patients tested when these studies were considered alone (p = 0.014; median = 27 patients in non-associations and 34 in association; n = 75 and 18 studies respectively). Patient numbers were not associated with studies concerned with PK or PFS/response (p > 0.66). No association was detected when the number of variants tested was compared to studies that demonstrated a statistical finding (p = 0.61; Wilcoxon rank sum test). However, numerous genes were studied, which likely confounded the analysis. The limited number of studies per gene prevented analysis of the number of variants tested within specific genes. The number of dose levels was also not associated with the detection of a statistical finding (p = 0.088; Wilcoxon rank sum test). Lastly, between 31 and 50% of studies on major phase I trial endpoints failed to provide any statistical analysis, typically due to low genetic variability or low patient numbers precluding an analysis.To our knowledge, the present analysis is the first to assemble and analyze several aspects of all published phase I clinical trials including pharmacogenetics in oncology. It is consistent with previous suggestions that pharmacogenomics assessments may need to be delayed for better powered phase II or III clinical trials in most circumstances [3]. Additionally, the endpoints of phase I studies examined in this review are a function of many factors that may reduce the penetrance of each genetic variant, such as age, race, sex, polypharmacy, prior therapy, and other factors [17]. Rarely are these factors included in multivariate analyses along with genotype despite heterogeneous patient cohorts in spite of a high degree of heterogeneity found in phase I trial designs. Most of these studies also focused on genes that were known to affect ADME-A or pharmacodynamics pathways even though tested variants in these genes did not have a high degree of analytic or clinical validity. Of those that did study well-characterized variants, almost none had sufficient coverage of important pharmacogenetic variants that are known to affect drug disposition. Lastly, it is understandable that pharmacogenetics is often a secondary endpoint of phase I studies, leading to insufficient recruitment to conduct a formal statistical analysis. Low genotype representation, however, can be overcome by including estimates of minor allele frequency in study design, recruiting racial populations in which pharmacogenetic variants are commonly inherited, or including genotyping in inclusion criteria.It is estimated that variation in genes that affect the pharmacokinetics or pharmacodynamics of medications accounts for approximately 20–30% of drug response variability overall [18]. To account for such variation during drug development, future phase I trials with pharmacogenetics endpoints should ensure that they are conducted with sufficient statistical power and a high degree of preclinical or clinical evidence, leveraging current knowledge about gene function prior to embarking on pharmacogenetics testing.Most genotype-directed dosing studies have tested differential dosing of irinotecan or other SN-38-related medications in patients carrying UGT1A1 variants [19,20,21,22,23]. Differential dosing for SN-38 was recommended in all studies. Other studies determine the capecitabine dose in patients carrying the 3R/3R genotype in thymidylate synthase (TYMS) were useful for capecitabine dosing [24], the dose of ocaratuzumab in patients carrying FC-gamma receptor IIIa (FCGR3A) variants [25], or whether batracyclin could be administered to those carrying slow acetylator phenotypes in N-acetyl transferase 2 (NAT2) in order to ensure low plasma concentration of a toxic metabolite [26]. In every case, these studies had a wealth of preclinical and/or prior clinical evidence to justify attempts to stratify dosing based on genotype [26,27,28,29].All genotype-directed Phase I studies in irinotecan only examined UGT1A1*28, a polymorphism in the UGT1A1 promoter that alters the length of a critical TATA box. Yet, there are four different possibilities of TATA box repeat length that are associated with decreasing levels of UGT1A1 expression at UGT1A1 (TA)n (rs3064744): (TA)5 (UGT1A1*36), (TA)6 (UGT1A1*1), (TA)7 (UGT1A1*28), and (TA)8 (UGT1A1*37) [30,31]. These variants are also detected with a variety of methods in phase I studies, including fragment sizing, pyrosequencing, PAGE gel sizing, or undisclosed methodology. However, we have demonstrated that many of these methods lead to incorrect UGT1A1 genotyping at this locus, calling the results of many of these studies into question. Decreased UGT1A1 function is also associated with UGT1A1*6 and UGT1A1*27 [32,33,34,35,36], which were not tested in these studies.Study design complications are also apparent in other genotype-directed studies. For example, the study examining TYMS genotyping examined the TYMS gene enhancer region (TSER) 2R/3R (rs45445694) and slow accrual resulted in only 5 patients with TSER 2R/2R + 2R/3R genotypes being recruited before this arm of the study was closed. Thus, no dosing guidelines were provided for this group of patients, and only one adverse event was reported [24]. Moreover, this study did not probe a well-characterized cysteine substitution in TYMS (rs2853542), nor did it evaluate an insertion/deletion polymorphism in the 3′ UTR (rs16430) that is associated with reduced TYMS transcription [28]. Patients who harbored the TSER 3R genotype may have then been treated at standard dosing in the presence of other allelic variants that may have influenced pharmacokinetics and toxicity. Thus, even though genotype-directed studies are better powered to answer scientific questions about gene–drug interactions, they too may be confounded by inaccurate and/or incomplete genotyping and limited statistical power.A total of seventeen phase I studies have been published examining irinotecan pharmacogenetics, although several studies compared multiple endpoints to genotype. Every one of these studies includes UGT1A1*28, although several other UGT1A1 alleles have been studied: UGT1A1*6, UGT1A1*27, UGT1A1*36, UGT1A1*37, and UGT1A1*60. Eight of these studies did not offer a formal statistical analysis, and eight other studies found no relationship between UGT1A1 alleles and pharmacokinetics (n = 2), toxicity (n = 2), response (n = 1), disease progression (n = 2), or survival (n = 1). Two studies found UGT1A1*28 was associated with inter-individual variation in pharmacokinetics [37,38] and two did not [39,40]. Three studies found UGT1A1*6 and/or UGT1A1*28 were associated with toxicity [37,40,41] and two did not [38,42]. No relationship between response or survival and any genotype was determined [42,43]. Others have evaluated variants in ABCB1, CYP3A4, CYP3A5, UGT1A6, UGT1A7, and UGT1A9; however, only one study found UGT1A6 phenotype status was related to toxicity [37]. As stated previously, some Phase I studies have studied differential dosing in patients with different UGT1A1 allelic variants [19,20,21,22,23]. Eight studies provided no formal statistical analysis for an association between UGT1A1 genotypes and clinical data derived from phase I studies [19,44,45,46,47,48].Despite several studies evaluating pharmacogenetic variants in anthracyclines [49], only two studies have evaluated the influence such variants on the toxicity and response in this class of agents. The first study evaluated amrubicin, finding no evidence that a single variant in NQO1 (609C > T) influences toxicity or response [50]. No formal statistical analysis was conducted for another study that evaluated SNPs in ABCB4, ABCC1, CBRR1, CBR3, FMAO2, HNMT, SLC10A2, SLC28A3, and UGT1A6 in relation to doxorubicin toxicity [51].One study tested two camptothecin derivatives (9-amino-camptothecin and 9-nitro-camptothecin) in a phase I study that compared variants in efflux transporters in relation to pharmacokinetics and toxicity. This study found that a variant in ABCG2 (Q141K; rs2231142) was associated with aminocamptothecin dose-normalized AUC but not toxicity [52]. A study of topotecan found no relationship between variants in CYP3A4, CYP3A5, UGT1A1, ABCG2, and ABCB1 and topotecan pharmacokinetics [53]. A study evaluating UGT1A1*28 and TAS-103 pharmacokinetics did not conduct a formal statistical analysis [54]. A genotype-directed dosing study in NAT2 slow acetylators was conducted for batracyclin, a topoisomerase I/II inhibitor. A dose was selected for NAT2 slow acetylators, who are at lower risk of exposure to a toxic batracyclin metabolite [26]. Lastly, one study evaluated several variants in drug metabolizing enzymes and AOX1 in relation to TP300 treatment, but this study offered no formal statistical comparison [55].Five studies have evaluated capecitabine toxicity and response, one of which also evaluated genotype-directed dosing. A polymorphism in CDA (79A > C) was associated with the development of hematologic toxicity in one study and diarrhea in another [56,57]. These studies also examined variants in DPYD, ENOSF1, GSTP1, MTHFR, and TYMS with no statistical differences in the development of capecitabine toxicity. Another study tested whether variants in CDA, DPYD, GSTP1, and TYMS were associated with capecitabine response in patients with anal cancer, finding no relationship [56]. Two studies evaluated MTHFR and TYMS variants in patients treated with 5-FU with no formal statistical analysis offered [58,59]. A single genotype-directed study evaluated differential dosing of capecitabine in patients with variants in TSER 2R/3R genotypes, as was mentioned previously [24].Pemetrexed pharmacogenomics has been frequently studied in the Phase I setting. Three studies evaluated variants in FPGS, GGH, GIF, MTHFR, SLC19A1, and TYMS in relation to pemetrexed toxicity and response, finding no relationships [60,61,62]. Conflicting evidence for a relationship between MTHFR 1298A > C (rs1801131) and disease progression or overall survival on pemetrexed in head and neck cancer or various solid tumors has been presented [60,61]. No relationship was found for other variants in MTHFR and TYMS in these studies.Ralitrexed and pralatrexate are poorly studied. A single study examined the MTHFR 667C > T (rs1801133) in relation to ralitrexed toxicity, finding that this variant was associated with overall toxicity [63]. Another study evaluated this variant, MTHFR 1298A > C, and the TYMS 2R/3R repeat polymorphism (rs45445694) in relation to pralatrexate toxicity, finding no relationship [64].Three studies have focused on gemcitabine therapy in the phase I setting. One evaluated LY2334737 toxicity, finding that SNPs in CDA (rs818202) and the HLA complex (rs3096691) were associated with the development of hepatotoxicity [65]. The other two studies either did not disclose the specific variants in the genes they tested [66] or did not conduct a formal statistical analysis [67].S-1 is an oral fluoropyrimidine that combines tegafur with a DPYD inhibitor, 5-cholor-2,24-dihydroxypyridine, and an orotate phosphoribosyl transferase inhibitor, potassium oxonate [68]. A single study evaluated CYP2A6 variants in relation to S-1 pharmacokinetics, finding that CYP2A6*4, *7 and *9 were associated with a lower metabolic ratio of S-1 (i.e., the exposure ratio of 5-FU to tegafur) [39].OSI-7904L is a liposomal formulation of a thymidylate synthase inhibitor that noncompetitively inhibits thymine nucleotide synthesis [69]. Two studies examined the TYMS 2R/3R repeat (rs45445694) and/or the 3R G/C (rs45445694) variant and found no association with these variants and OSI-7904L toxicity or response [70,71]. A third study detected the same polymorphisms in addition to MTHFR 677C > T (rs1081133) but did not conduct a formal statistical analysis [69].Six studies have evaluated whether pharmacogenomics influences Phase I studies of antiangiogenesis agents. A single study evaluated whether variants in three drug efflux transporters were associated with telatinib pharmacokinetics and whether variants in FLT4 and VEGFR2 were associated with the development telatinib toxicity; however, no association was detected [72]. Another study found a variant in VEGFA (rs833061) was associated with the development of high-grade neutropenia in those treated with pazopanib [62]. Another study evaluating pazopanib pharmacogenetics found CYP3A4*22 carriers had lower pazopanib clearance, whereas variants in ABCB1, and ABCG2 were not related to pazopanib PK [73]. Progression and overall survival following sorafenib has also been examined in the Phase I setting for those with various solid tumors or pancreatic cancer [74,75]. A variant in VEGFA (-899GG) was associated with PFS of sorafenib, and two variants were associated with OS (-1154AA and -7TT), although not consistently between the two studies. Two other studies genotyped a wide variety of SNPs in several genes with a possible relationship to vatalanib or pazopanib pharmacology, but neither study conducted a formal statistical analysis [76,77]. Two studies evaluated bevacizumab response or PFS: The first study found that PFS duration was shorter in those carrying the rs6900017 genotype [78], and the second did not provide a formal statistical analysis of VEGFA genotypes versus response in patients treated with both bevacizumab and sorafenib [79].Topoisomerase inhibitors, antimetabolites, and antiangiogenic agents represent 116 of the 206 total comparisons and 49 of 82 studies covered in the present review. Multiple lines of evidence suggest that variants in UGT1A1 are strong predictors of SN-38 metabolism, pathway variants in folate metabolism (i.e., TYMS and MTHFR) are commonly associated with antimetabolite therapy efficacy, and pathway variants in angiogenesis affect several VEGFA and VEGFR2 (KDR) inhibitors [27,28,80]. It is not surprising that over half of phase I studies account for variants in these genes. Yet, there is no statistical relationship between the number of studies detecting an association with pharmacogenetic variants in the above studies (22 comparisons detected an association and 57 did not) versus those devoted to testing other gene–drug interactions (18 comparisons detected an association and 39 did not; p = 0.70; Fisher’s exact test). Again, phase I studies may not be the best platform to answer scientific questions about the relationship between pharmacogenetic variants and outcomes.While many of these phase I trials covered herein were conducted prior to the characterization of the analytical or clinical validation of pharmacogenetic variants, the present review clarifies that even modern phase I studies have design complications that frequently preclude or seriously limit answering scientific questions about inter-individual variability attributed to genetics. The goal of phase I trials is to find a safe dose for phase II studies while simultaneously understanding the pharmacologic and PK properties of agents in humans. While assessment of response is not the goal, many phase I studies try to detect a response signal. Except for studies of molecularly targeted agents, phase I studies in oncology attempt to define the maximum tolerated dose of anticancer agents to maximize the potential for response with acceptable toxicity, resulting in a narrow therapeutic window in which inter-individual variation in toxicity or pharmacokinetics can seriously influence outcomes. Thus, early patient stratification can increase success during early development and is desirable from the standpoints of patient safety, increasing efficacy rates, and mitigating the attrition rate of drug development in oncology.Phase I trials, however, are not restricted to homogeneous populations with different diseases, prior therapies, comorbidities, and other factors that confound statistical relationships in gene–drug interactions. The majority of phase I studies included herein also included combinations of various medications (48 of 84 studies) that may further confound statistical analysis, and many of them fail to conduct a statistical analysis. Such heterogeneity in small patient populations does not lend itself to hypothesis-free genotyping methods; thus, it is not surprising that Phase I studies most commonly use candidate gene methods. However, coverage of genetic variants is also poor in most of these trials. While small studies often need to avoid multiple comparisons, many of these studies may be confounded by unstudied genetic variation—particularly in genes for which several variants are known to influence gene activity. This detraction of phase I studies is simple to correct by studying activating or deactivating variants to inform gene activity in several genes for which this information is readily available. Multigene technologies, such as Pharmacoscan (formerly the DMET array; Thermo Fisher Scientific), probe multiple variants in well-characterized pharmacogenes and classify these variants into a set of curated phenotypes, but such methods were only used in one study we evaluated [13]. Candidate genes often have poor preclinical or clinical justification for testing in the clinical setting, and candidate gene variants frequently have low analytical/clinical validity in phase I studies. Overall, far fewer than 1% of phase I trials include pharmacogenetics (see methods section). Accounting for these difficulties during study design may make pharmacogenetics testing in phase I studies more routine. Moreover, as the cost for developing oncology agents approximates $2.8 billion United States dollars [85], the expense of early testing of genetic variation is miniscule. Thus, appropriately designed pharmacogenetics testing will likely provide a significant return on significant time and investment required to move oncology agents into humans.Conceptualization, T.M.S. and W.D.F.; methodology, T.M.S. and W.D.F.; formal analysis, T.M.S.; investigation, T.M.S.; resources, W.D.F.; data curation, T.M.S.; writing—original draft preparation, T.M.S. and W.D.F.; writing—review and editing, T.M.S. and W.D.F.; funding acquisition, W.D.F. All authors have read and agreed to the published version of the manuscript.This research was funded by National Institutes of Health intramural funds grant number [ZIA BC 010627].The authors declare no conflict of interest.The views expressed here are those of the authors and do not necessarily reflect the views of the National Cancer Institute, the National Institutes of Health, the Department of Health and Human Services, or the United States government.Polymorphic metabolic enzymes affect pharmacokinetics and pharmacodynamics of medications by activating/inactivating them and encouraging their elimination. Transporters similarly affect pharmacokinetics and pharmacodynamics by encouraging or preventing distribution of compounds to or from bodily compartments. Some studies examine how genetic variation affects medications at their site of pharmacologic action by studying direct or indirect effects of drug action on biological pathways.Important parameters of phase I studies incorporating pharmacogenomics approaches.Phase I study design factors categorized by gene-drug pairs and study endpoint.* Genotype-directed study.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01132.txt

@@ -0,0 +1 @@
+High-grade serous ovarian cancer (HGSOC) is the most frequent and lethal form of ovarian cancer and is associated with homologous recombination deficiency (HRD) in 50% of cases. This specific alteration is associated with sensitivity to PARP inhibitors (PARPis). Despite vast prognostic improvements due to PARPis, current molecular assays assessing HRD status suffer from several limitations, and there is an urgent need for a more accurate evaluation. In these companion reviews (Part 1: Technical considerations; Part 2: Medical perspectives), we develop an integrative review to provide physicians and researchers involved in HGSOC management with a holistic perspective, from translational research to clinical applications.High-grade serous ovarian cancer (HGSOC), the most frequent and lethal form of ovarian cancer, exhibits homologous recombination deficiency (HRD) in 50% of cases. In addition to mutations in BRCA1 and BRCA2, which are the best known thus far, defects can also be caused by diverse alterations to homologous recombination-related genes or epigenetic patterns. HRD leads to genomic instability (genomic scars) and is associated with PARP inhibitor (PARPi) sensitivity. HRD is currently assessed through BRCA1/2 analysis, which produces a genomic instability score (GIS). However, despite substantial clinical achievements, FDA-approved companion diagnostics (CDx) based on GISs have important limitations. Indeed, despite the use of GIS in clinical practice, the relevance of such assays remains controversial. Although international guidelines include companion diagnostics as part of HGSOC frontline management, they also underscore the need for more powerful and alternative approaches for assessing patient eligibility to PARP inhibitors. In these companion reviews, we review and present evidence to date regarding HRD definitions, achievements and limitations in HGSOC. Part 1 is dedicated to technical considerations and proposed perspectives that could lead to a more comprehensive and dynamic assessment of HR, while Part 2 provides a more integrated approach for clinicians.High-grade serous ovarian cancer (HGSOC) is the most frequent and lethal form of epithelial ovarian cancer (EOC) [1]. Despite substantial improvement in the clinical management of HGSOC, the all-stage 5-year overall survival (OS) rate remains at approximately 40% [2,3,4]. Thus, a better understanding of this disease is urgently required, from molecular deciphering to new therapeutic molecules. As such, a novel class of molecules, called polyadenosine diphosphate-ribose polymerase (PARP) inhibitors, emerged during the last decade. PARP inhibitors (PARPis) are based on homologous recombination deficiency (HRD), a molecular alteration that affects approximately 50% of HGSOC cases. In parallel with PARPi development, HRD assays have been developed to provide clinicians with accurate estimates of homologous recombination status. As such, PARPis, coupled with HRD assays, led to substantial improvements in HGSOC prognosis [5].However, HRD assays remain controversial, notably due to their technical and medical relevance [6,7]. By bridging the gap between molecular and clinical considerations, these companion reviews will present the evidence to date regarding HRD definitions, achievements and limitations in EOC, with the aim of providing physicians and researchers involved in HGSOC management with a holistic perspective, from translational research to clinics. Part 1 focuses on molecular and technical considerations, describing: 1. the main components of HRD through a dichotomic approach (i.e., causes and consequences); 2. the rationale, development and technical performance of current HRD assays; and 3. the limitations inherent to current HRD assays and the axes of research and proposed perspectives that could lead to a more comprehensive and dynamic assessment of HRD, with the aim of improving its predictive value. The companion paper (Part 2) focuses on clinical considerations and, notably, the impact of PARPis in the clinic.Discoveries in observational studies and family studies led to the determination of genetic factors involved in EOC risk, notably those associated with the so-called hereditary breast and ovarian cancer (HBOC) syndrome caused by germline mutations in BRCA1 or BRCA2 (BRCA1/2). Based on The Cancer Genome Atlas (TCGA) database, it is estimated that 15–20% of HGSOC patients carry a germline mutation in BRCA1/2 genes, leading to a cumulative risk of developing EOC by the age of 80 years of 44% and 17%, respectively, versus 1.4% in the general population [8,9]. BRCA1 mutation carriers exhibit OC at a younger age than BRCA2 mutation carriers. The high prevalence of BRCA1/2 mutations led to the formulation of international guidelines to integrate genetic testing, or at least genetic counseling, upon EOC diagnosis, particularly in the context of a familial history of OC [10,11,12,13]. Furthermore, prophylactic risk-reducing bilateral salpingo-oophorectomy has been shown to be an effective prevention strategy in germline BRCA1/2 carriers [14]. It also should be noted that BRCA2 germline mutation increases the risk of prostate and pancreatic cancers [8,13]. A few cases of constitutive epimutations (i.e., aberrant hypermethylation) of the BRCA1 promoter have been reported [15,16]. Other germline mutations have also been described, such as RAD51C, RAD51D, PALB2, BARD1 or BRIP1, representing a cumulative frequency of approximately 5% [17,18]. Notably, inherited mutations in mismatch repair genes (mainly hMLH1, hMSH2, hMSH6 and PMS2), which lead to Lynch syndrome, are also known to increase the risk of OC, although they mainly exhibit endometrioid histology [19]. Beyond germline alterations, EOC depends on a few crucial pathways. Indeed, TP53 (which encodes the tumor suppressor p53) is mutated in approximately 96% of HGSOC cases and is considered an early event and driver mutation of cancer progression [20,21]. Data from TCGA have shown recurrent mutations in a restricted set of genes: BRCA1, BRCA2, NF1, RB1 and CDK12 [9]. Furthermore, HGSOC is also characterized by frequent chromosomal instability through DNA gain/loss, leading to tumor suppressor gene (TSG) loss and oncogene amplification [22].Strikingly, several genes altered in HGSOC are involved in DNA repair through the homologous recombination (HR) process. Briefly, single-strand breaks (SSB) are processed through the base excision repair (BER) mechanism, mainly involving PARP proteins (of whom PARP1 is the most characterized). When facing double-strand break (DSB), cells will use different repair mechanisms: homologous recombination (HR), non-homologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ). While the HR process leads to faithful DNA repair by using a homologous template, end resects are directly ligated by NHEJ, potentially leading to small insertions/deletions (indels). MMEJ also leads to specific indels, which are longer than the ones occurring during NHEJ [23].At the genetic level, alterations to BRCA1 and BRCA2 are the most frequent and best -characterized alterations in the HR pathway. Indeed, both germline and somatic deleterious mutations in BRCA1/2 genes (referred to as gBRCA* and sBRCA*, respectively) have been shown to promote HGSOC [9,24]. BRCA1/2 play key roles in genome integrity maintenance and their alteration are an early event in the EOC carcinogenetic process; indeed, the loss of the first allele of BRCA1 (or BRCA2) is a facilitating event for TP53 loss, both of them leading to EOC development [25]. The most frequent and well-known alterations are short mutations in BRCA1/2 genes, leading to coding sequence disruption (through missense, nonsense or frameshift mutations) that subsequently inactivate proteins or result in dominant-negative mutations [26]. According to the two-hit Knudson hypothesis, in the context of gBRCA*, all the cells in the patient already carry an inactive copy of BRCA (the first hit); thus, the loss of the second allele (the second hit), which mainly occurs through loss of heterozygosity (LOH), is the only step needed to produce HRD [27]. In contrast, patients without gBRCA* need two hits to develop HRD. gBRCA1* and gBRCA2* are estimated to occur in 8% and 6% of HGSOC cases, respectively. Furthermore, sBRCA1* and sBRCA2* are found in an additional 4% and 3% of cases, respectively [9,28]. Notably, it has been suggested that BRCA1 haploinsufficiency due to gBRCA1* could be sufficient to initiate tumorigenesis through the TP53 mutation, without needing a second hit [29]. This hypothesis is supported by the fact that retention of the normal allele (i.e., absence of LOH) has been reported to occur in 7% and 16% of HGSOC cases that carry gBRCA1* or gBRCA2*, respectively [30]. Nevertheless, in the context of sBRCA*, LOH is considered to be near-universal; additionally, a few studies have reported that short mutations could be an alternative second hit [31,32].Recently, multimegabase large rearrangements (LRs) at the BRCA1/2 loci have been suggested to lead to HRD, accounting for approximately 16% of patients [33]. Owing to their different structures, the BRCA1 and BRCA2 proteins exhibit distinct functions in the cell. Aside from HR, in which the role of BRCA2 remains unclear, BRCA1 exhibits a variety of functions in the cell, such as cell cycle regulation, chromatin remodeling, replication fork protection and apoptosis [34,35].In addition to direct alterations to DNA sequences, epigenetic mechanisms can also lead to loss of expression of BRCA1. Indeed, BRCA1 promoter methylation (-CpG+) has been characterized for decades [36]. Data from TCGA revealed that the BRCA1-CpG+ mechanism leads to HRD in approximately 11% of HGSOC cases [9]. According to a meta-analysis based on 16 studies, BRCA1-CpG+ is found in approximately 16% of EOCs [37]. A recent study, which analyzed 88 EOCs, found BRCA1-CpG+ and BRCA2-CpG+ in 19.3% and 4.6% of the cases, respectively [38]. This latter finding could imply that BRCA2-CpG+ is a cause of HRD, although it has been classically considered a quite rare event. Interestingly, BRCA mutations and BRCA1-CpG+ are almost mutually exclusive [9].Although initially described with BRCA1/2*, HRD was later characterized in the context of wild-type BRCA (BRCAwt); therefore, the so-called “BRCAness” phenotype, encompassing any HRD not caused by a direct BRCA alteration, was identified [39]. In BRCAwt, biallelic mutations in HR genes that lead to BRCAness have been described. They represent approximately 5% of cases and mainly concern mutations to RAD51C, RAD51D, BRIP1 and PALB2 [40,41,42]. In addition to point mutations, LRs that affect genes other than BRCA1/2, such as RAD50 and NBS1, have been reported [43]. Furthermore, HRD can be the consequence of EMSY (a BRCA2-interacting transcriptional repressor) amplification, an alteration found in approximately 5% of cases [44,45]. Notably, a specific subset of HGSOC cases exhibits CCNE1 amplification, a molecular carcinogenetic pathway associated with HR proficiency and a poor prognosis [46].In addition to BRCAness due to genetic mutations, promoter methylation in other HR genes, such as RAD51C and PALB2, has been described in HGSOC [26,47]. Apart from protein-coding RNAs, the role that microRNAs (miRNAs), typically studied in the regulation of gene expression through translation inhibition and mRNA degradation, play in the carcinogenetic process has begun to emerge, with specific miRNA signatures associated with OC [48,49]. Deregulation of some miRNAs, such as miR-509-3p and miR-211, through HR inhibition, has been implicated in platinum-sensitive (Pt-S) cancers [50,51]. Long noncoding RNAs (lncRNAs), which exhibit a wide range of physiological functions, such as transcription regulation and scaffolding within cells, have been linked to OC risk, carcinogenesis and prognosis [52,53]. For instance, PCAT-1 leads to HRD in prostate cancer by suppressing BRCA2 [54]. Although several lncRNAs have been shown to be involved in HR, their actual etiologic impact on HRD still needs to be assessed in EOC [55]. Interestingly, it has been suggested that miRNA and lncRNAs deregulations are caused by alterations in methylation, implying crosstalk between these two epigenetic processes.Chromatin dynamics rely mainly on posttranslational modifications (PTMs), nucleosome positioning and spatial genome organization; these modifications are involved in complex crosstalk with DNA methylation [56]. Currently, there is no clinical proof that such epigenetic modifications can directly lead to HRD in the context of OC [57]. However, as chromatin can act as a gene silencer and a barrier to efficient DNA repair, it appears plausible that chromatin remodeling could partly explain (or at least participate in) HRD in OC [58]. Indeed, several histone PTMs have been described in the context of DSBs, notably participating in the selection between the HR and NHEJ pathways [59,60]. Interestingly, H2AX phosphorylation at serine 139 (the so-called γ-H2AX) is considered the hallmark of DSBs [61]. H4K16 acetylation and H3K36 trimethylation act synergistically and are required for HR after a DSB; loss of these markers leads to inefficient HR [62,63]. Bromodomain-containing protein 9 (BRD9), which has been shown to be mutated in EOC, is essential for acetylation of RAD54 and its interaction with RAD54 is likewise necessary for efficient HR [64].In conclusion, although not easily translatable to routine clinical practice, many of the intricate epigenetic mechanisms involved in HR and its deficiency have been described thus far.Owing to the democratization of next-generation sequencing (NGS), intensive oncological research has been conducted to investigate correlations between specific cancer types and distinct sets of DNA alterations (the “mutation signature”). In addition to histomolecular correlations, data on specific carcinogens (e.g., tobacco smoke) and tumor aggressiveness have led to molecular classifications [65,66]. Thus, the consequences of HRD can be classified into three different categories: DNA alterations, epigenetic markers and functional phenotypes (Figure 1).DNA alterations can be divided into two classes: microlesions (affecting a single to a few nucleotides) and macrolesions (affecting larger DNA regions, up to megabases). At the microlesional scale, BRCA alterations (caused by gBRCA1/2*, sBRCA1/2* or BRCA1-CpG+) lead to a specific base substitution pattern, named “signature 3” [67,68,69]. In contrast to the other defined signatures, the BRCA1/2*-associated signature exhibits a relatively equal distribution of different base substitutions (i.e., transitions and transversions) that are quite homogeneous across the gene. Furthermore, it is characterized by a high number of microhomology-mediated deletions, which are consequences of HRD and the compensatory use of the MMEJ mechanism. Interestingly, BRCA1/2* leads to a seven-fold increase in base mutagenesis through HRD, thus promoting oncogenesis and intratumoral heterogeneity [70]. In addition to BRCA alterations, signature 3 has been observed in other HR-related genes, leading to a BRCAness molecular signature [71,72,73]. Notably, this signature is not present in the context of incomplete inactivation of HR-related genes [35].At the (sub)chromosomal scale, HRD leads to gross rearrangements and aberrations [74]. Several types of LR have been described as consequences of HRD. Branded with the generic term “genomic scars”, they constitute a permanent fingerprint of HRD-related global genomic instability [75]. The literature tends to use “mutational signatures” and “genomic scars” to refer to microlesions and macrolesions, respectively; however, these are interchangeable. HRD leads to a specific panel of copy number alterations through deletions, duplications, inversions and translocations. Indeed, BRCA1/2* breast tumors exhibit a specific genomic profile in array comparative genomic hybridization (aCGH), a profile that can also be found in tumors that exhibit BRCAness [76,77].More precisely, three types of alterations are enriched in tumors with HRD: LOH, large-scale state transitions (LSTs) and telomere allelic imbalance (TAI). Stricto sensu, LOH is an allelic imbalance (i.e., loss of equilibrium between paternal and maternal alleles); it can be either “copy defective” (i.e., a simple loss of one of the alleles and subsequent haploidy in a given locus) or “copy neutral” (i.e., no loss of diploidy) [78]. In the context of HRD, LOH refers to deletions >15 megabases (Mb) but less than a whole chromosome [79]. TAI is defined as an allelic imbalance >11 Mb in subtelomeric regions [80]. Conversely, LST refers to an allelic imbalance >10 Mb in size between adjacent genomic regions, a phenomenon caused by translocations and insertions/deletions [81]. Notably, in the context of HRD, these alterations are found across the genome, and their global enrichment is reflected in the global genomic instability score (GIS) [82].In addition to genetic alterations, HRD is also associated with epigenetic alterations. However, the link between cause and consequence is harder to establish for epigenetic alterations. Indeed, the different HRD-related epigenetic alterations can precede, maintain or occur after an HRD phenotype develops.The simplest layer of epigenetics is based on specific gene expression profiles (GEPs), secondary to DNA mutations [83]. Most of the studies assessing epigenetics to date have been performed in the context of BRCA*. Seminal studies have shown specific GEPs are associated with BRCA* tumors, with a distinct panel depending on the gene affected (BRCA1* versus BRCA2*) and on the etiology (germinal versus sporadic); intriguingly, a partial overlap exists among these and a subset of BRCAwt HGSOC cases, suggesting a common HRD-associated GEP [84,85,86]. A 60-gene-specific panel associated with BRCAness has been described [87]. Specifically, HRD exhibits a distinct core GEP [88]. Notably, in the context of gBRCA1*, even healthy fallopian tissue exhibits a specific GEP, suggesting an epigenome-modifying influence [89,90]. The partial divergence observed between BRCA1* and BRCA2* tumors probably relies on the pleiotropic function of BRCA1, which is not restricted to DSB management [91,92].Several epigenetic markers have been associated with HRD. Importantly, BRCA1 has been shown to negatively regulate Polycomb-repressive complex 2, a major chromatin remodeling enzyme involved in stem cell-state maintenance through transcriptional repression of histone H3K27 trimethylation [93,94]. Following HRD, tumors exhibit specific PTMs, with lower levels of H4K12/16-acetylation and overexpression of histone deacetylase 6 [95,96]. In patients with gBRCA1/2*, apparently “normal” fallopian tissue still carries a reprogrammed epigenome with a specific methylome [97]. The inefficiency of HR leads to a lack of RAD51 recruitment at DSB sites (the so-called “RAD51 foci”) [98,99]. Furthermore, a subnetwork of 30 co-expressed proteins distinguishes HRD versus non-HRD HGSOC [96].At the functional level, HRD tumors are metabolically distinct; indeed, they rely more on oxidative phosphorylation than glycolysis [100]. Owing to error-prone DSB processing, HRD tumors have a higher tumor mutational burden (TMB), neoantigen load and HLA-I expression [101]. This is in accordance with the increase in tumor-infiltrating CD8+ lymphocytes and the higher expression of PD-1/PD-L1 proteins observed in BRCA1/2* HGSOC, reflecting an “immunologically hot” phenotype [102].Clinically, HRD tumors tend to be more sensitive to platinum-based regimens and PARPis, although differences exist depending on the underlying HRD mechanism [98,103]. Platinum-based treatments mainly rely on DSB generation via crosslinks with DNA, a phenomenon that is highly toxic to cells in the context of HRD [104]. In terms of PARPis, the sensitivity of HRD tumors is based on their synthetic lethality (SL). This concept relies on the fact that cancer cells harbor gene defects that are not lethal per se, but that become lethal when combined with a defect in another gene [105]. When using PARPis, SL occurs because of the inability of HRD tumors to manage DSBs. Some PARPs (a large family of 17 proteins that participates in several cellular pathways through the ADP-ribose PTM and whose deregulation is implicated in carcinogenesis) are involved in DNA repair. PARP1, which is the most characterized PARP, plays a key role in SSB repair mainly through BER, although its role in MMEJ has recently been described [106,107]. Consequently, PARPis impairs the BER pathway, leading to SSB accumulation and progression to DSBs. In HR-proficient cells, these DSBs will be processed, allowing continued cell viability. In contrast, HRD cells accumulate DSBs, ultimately leading to apoptosis. Interestingly, recent studies have shown that PARPis have other roles, such as stalling the replication fork, stalling or trapping PARP1 on DNA (leading to protein-DNA adducts) and subsequent cell death in HRD cells [108]. In the context of HGSOC, seminal studies showed both in vivo and in vitro a specific SL occurring with PARPis in a BRCA-deficient context [109,110]. Subsequently, this breakthrough class of agents started to emerge in randomized clinical trials with substantial improvements in patients with HRD HGSOC, leading to approval of three molecules to date: olaparib, niraparib and rucaparib. Clinical considerations will be developed in Part 2 of this review. Noteworthy, HRD HGSOC spontaneously tends to have improved progression-free survival (PFS) and OS, which is in part due to better treatment responses [111].Consequently, an accurate evaluation of the HRD status of cases remains essential, for both prognosis and theranostics.Owing to the potent impact of HRD status on HGSOC management, several assays have been developed for HRD evaluation; while some have been used in research thus far, others are currently used as “companion diagnostic” (CDx) assays prior to PARPi prescription. This section will focus on clinically validated and/or routinely used tests (Table 1), while tests currently under investigation will be discussed in the fourth part of this review. This review will only focus on technical considerations (performances and limitations of each test), while the clinical considerations (i.e., the relevance of evaluating HRD status as a biomarker for PARPi prescription and response) will be detailed in the related paper. To date, three CDx assays have currently received FDA approval for OC [112]. HRD evaluation mainly relies on two strategies (following the cause versus consequences dichotomy): searching for mutations in HR-related genes (mainly BRCA*; the causes of HRD) and/or the presence of “genomic scars” (the consequences of HRD).Currently, the mutation statuses of BRCA1 and BRCA2 can be either evaluated through gene-specific (i.e., BRCA1/2 targeted sequencing) or multipanel testing, the latter detecting potently targetable non-BRCA alterations. Based on central sequencing confirmation, the gBRCA tests provided by different companies have been shown to produce reliable results concerning molecular alterations, with a concordance of >95% [113]. The rate of variants of unknown significance (VUS), which was originally 84% during initial BRCA testing development, has dropped to approximately 10% due to VUS reclassification and refinement, notably through the Consortium of Investigators of Modifiers of BRCA (CIMBA) [114]. However, non-BRCA HR-related genes, when used in some randomized controlled trials (RCTs) evaluating PARPis or Pt-S in HGSOC, have produced conflicting results and are currently under investigation [11,12,40,42].Depending on the context, BRCA analysis can be performed either in blood samples (i.e., constitutive) or directly within the tumor (tBRCA). While detection of gBRCA* in blood samples generally implies altered BRCA within the tumor, the presence of a mutation within the tumor (i.e., tBRCA*) from formalin-fixed paraffin-embedded tissue (FFPE) can be due either to sBRCA* or gBRCA*. Current international guidelines and practices recommend tBRCA* testing, at a minimum, for newly diagnosed HGSOC, though systematic gBRCA evaluation recommendations vary between guidelines [10,11,22,115,116].As gBRCA* represents the vast majority of tBRCA* and has important implications for the hereditary risk of breast and ovarian cancers (HBOC) and screening of patient relatives, current guidelines systematically recommend referral to genetic counseling upon tBRCA* detection [11,12,115,116]. Evaluation of gBRCA needs to be performed both at the micro- and macro-lesion scales, as LR constitutes a non-negligible fraction of BRCA alterations. Following the decision of the USA Supreme Court concerning the ineligibility of BRCA gene sequencing patents, many institutes and companies have developed their own assays [117].MyriadGenetics (MG), which initially characterized the sequence of BRCA, developed the BRACAnalysis® CDx in accordance with its extensive experience with BRACAnalysis® of HBOC [118]. In 2014, BRACAnalysis® CDx (BA-CDx) became the first FDA-approved CDx, along with olaparib, for treating patients with gBRCA* advanced OC who previously received ≥3 lines of chemotherapy [119]. BA-CDx relies on sequencing genomic DNA obtained from whole blood samples collected in EDTA. The entire coding sequences of the BRCA1/2 genes are included (as well as promoter regions and intron/exon boundaries). Point mutations and short indels are analyzed with polymerase chain reaction (PCR) and Sanger sequencing, while large rearrangements (LR-including deletions/duplications) are detected with multiplex PCR and a proprietary system entitled BART® CDx.There are five different classifications of results: positive for a deleterious mutation; genetic variant-suspected deleterious; genetic variant, favor polymorphism; genetic VUS; and no mutation detected. The VUS classification is the one of highest concern for geneticists and oncologists, with a moving interpretation according to scientific discoveries. MG uses its own private database, which relies on its extensive experience with BRCA analysis for determining hereditary cancer risk and variant classifications.To MG (consistent with the American College of Medical Genetics and Genomics recommendations), mutations are considered deleterious if they fall in one of the following categories: nonsense and frameshift mutations occurring at/or before the last known deleterious amino acid position of the affected gene; deletions/duplications of entire exons; LR leading to frameshifts; and mutations/LR based on the data derived from the linkage analysis of high-risk families, functional assays, biochemical evidence, statistical evidence, and/or demonstration of abnormal mRNA transcript processing [120]. MG classification is performed by a committee of experts (i.e., board-certified laboratory and medical directors, research scientists, genetic counselors and variant support specialists) and further enriches their proprietary database [121].According to the technical information given by MG, analytical validation studies (i.e., nonclinical studies) of BA-CDx were performed on a set of 110 samples containing single nucleotide variants (SNVs), deletions of up to 40 base pairs (bp) and insertions of up to 10 bp, with a validated NGS-based assay as comparator [121]. Agreement analyses included 100% positive percent agreement (PPA), negative percent agreement (NPA), and overall percent agreement (OPA). BART® CDx accuracy was evaluated by using a validated microarray assay on a set of 103 patients (with 29 samples positive for a large rearrangement in BRCA1 or BRCA2). BART® CDx yielded valid results for 98 samples, with concordance between the 2 tests for 97 samples (the discordant sample was identified as duplication by BART® CDx and triplication by the reference assay) [121].Clinical validation was performed on samples from several studies that evaluated the effect of PARPis on OC. Notably, a subset of 61 samples from Study 42 (NCT01078662), initially tested locally, was retrospectively tested with BA-CDx, and provided a 96.7% (59/61) concordance rate (the 2 discordant samples included 1 sample without a callable result with BA-CDx and another with a different variant classification result). Patients from the SOLO1 RCT (NCT01844986) enrolled either through prospectively testing with BA-CDx (n = 181) or local testing (n = 210). Of these 210 patients, 208 were retrospectively tested, and concordance was achieved for 207 of them (98.5%) [113,121].The main limitation of the BA-CDx assay is that it only detects gBRCA*; as gBRCA* represents approximately 70% of BRCA-mutated patients; a negative BA-CDx result does not rule out sBRCA*. Other limitations of the BA-CDx include the lack of general detection of insertions that differ from duplications and unequal allele amplification (and subsequent false-negative results) from rare polymorphisms under primer sites [121]. BA-CDx is currently FDA-approved as a CDx assay in three distinct situations where gBRCA* evaluation is required prior to PARPi prescription, as detailed in Table 1.Two commercially available CDx assays have been prospectively validated for the evaluation of genomic scars thus far: FoundationFocus CDxBRCA-LOH® (FF-CDx) from FoundationMedicine (FM) and MyChoice CDx® (MC-CDx) from MG [11,22]. Both tests combine tBRCA sequencing and genomic scarring evaluation. As BRCA analysis is performed on tumoral tissue, it does not distinguish between gBRCA* and tBRCA*.FF-CDx is based on comprehensive (i.e., including characterization of point mutations and indels) deep NGS, and thus has a >95% sensitivity and >99% positive predictive value; the starting material is FFPE, either in block form or on at least 10 unstained slides, with a minimum of 20% malignant tissue for BRCA1/2 analysis [122,123]. It relies on whole-genome shotgun library construction and hybridization-based capture, amplification and sequencing of the constructed library. Interestingly, while the capture process targets approximately 1.5 Mb of the human genome (including all coding exons of 310 cancer-related genes, introns or noncoding regions of 35 genes and >3500 SNVs located throughout the genome), the FF-CDx only reports results for BRCA1/2, raising the question of “lost data”. Subsequently, tBRCA1/2* (including SNVs and indels up to 13 bp) are detected through a custom analysis pipeline, with a 5% mutation allele frequency (MAF) cutoff (lowered to 1% for SNVs and 3% for indels mutations in hotspots). tBRCA* is considered deleterious when it leads to premature stop codons (PSCs) anywhere in BRCA1/2 coding regions (with the exception of the BRCA2 PSC at position K3326 and 3′ downstream), splice site alterations (defined as mutations at intron/exon junctions, ±2 bp from exon starts/ends) or deleterious missense alterations (according to the curated list based on the Breast Cancer Information Core database).An analytical validation study of FF-CDx was performed on a set of 36 tBRCA* (including SNVs, deletions up to 12 bp and insertions up to 4 bp) and 44 tBRCA wild-type (tBRCAwt) samples, with a validated NGS-based assay as a comparator. Agreement analyses were as follows: 100% PPA, 94.9% NPA and 97.3% OPA. The limits of detection (LODs, defined as the minimal allele frequency necessary to detect a given lesion) vary by type of alteration: 6% for SNVs and indels ≥12 bp in non-repetitive regions and 15.3% for deletions in homopolymer regions. Clinical validation was performed on a subset of the samples from Study 10 (NCT01482715) and the ARIEL2 (NCT01891344) study. The clinical bridging study, which compared FF-CDx versus local testing for BRCA1/2 evaluation, was performed on 67 samples and showed 97% PPA, 100% NPA and 97.9% OPA [124]. LOH was calculated through an almost genome-wide (i.e., the 22 pairs of autosomes) analysis of the >3500 SNVs detected in the hybridization-based capture process, leading to a global score that reflected the percentage of genomic LOH. Tumors were defined as “LOH high” (≥16%) or “LOH low” (<16%), corresponding to HRD-positive (HRD+) and HRD-negative (HRD-) statuses, respectively [125]. Notably, a positive HRD result was produced if the tumor was “LOH high” and/or exhibited a tBRCA*. Laboratory validation of the LOH component of the assay indicated the LOD was ≥35% DNA tumor content for a reliable analysis; furthermore, the accuracy of the LOH evaluation (calculated through inter-run reproducibility) was lower when the LOH value approached the 16% cutoff. Notably, “LOH high” initially had a different cutoff (≥14%), as it was defined through TCGA analysis for the ARIEL2 study [126].Interestingly, the FF-CDx evolved in parallel with RCTs that evaluated the PARPi rucaparib. Indeed, FoundationMedicine initially proposed a test entitled FoundationFocus CDxBRCA© that only focused on tBRCA alterations (point mutations and short indels); this test received FDA approval in 2016, along with the PARPi rucaparib, for patients with tBRCA*-associated advanced OC with ≥2 lines of chemotherapy [127]. This test subsequently included LOH analysis following the ARIEL2 and ARIEL3 (NCT01968213) studies that evaluated rucaparib, with the aim of determining the HRD status of patients with mutations beyond tBRCA*; the LOH analysis was performed with FM T5 NGS, which is an assay developed for clinical trials and has the same characteristics as the marketed FF-CDx [126,128]. Interestingly, 6% (ARIEL2) and 8.7% (ARIEL3) of the LOH evaluations provided inconclusive results.Importantly, FF-CDx is no longer available as a stand-alone assay but is provided within the more general FoundationOne CDx© (F1-CDx), driving a comprehensive multi cancer analysis. Indeed, beyond the tBRCA* and LOH evaluations, F1-CDx can detect mutations in a panel of 324 genes captured through the hybridization-based process, including copy number alterations, TMB, MSI and specific gene rearrangements [22]. The results are provided in three classes: CDx claims, cancer mutations with evidence of clinical significance, and cancer mutations with potential clinical significance. Comparison with the LOH evaluation of FF-CDx produced 97.5% PPA, 95.1% NPA and 96.7% OPA.According to its technical information, the F1-CDx assay allows LR identification; however, its concordance with other validated methods has not been evaluated. Consequently, confirmatory validation is required upon copy number alterations that affect BRCA (except for whole BRCA1/2 homozygous deletion). Furthermore, although this test theoretically has an LOD of >20% tumor purity for LR (importantly, this value is given for all HR pathway genes), the clinical bridging study using the SOLO1 (which only enrolled patients with tBRCA*) samples gave conflicting results. Indeed, 368 (94.1%) patients from SOLO1 were retrospectively tested, and 335 had a valid F1-CDx result. Of these 335 patients, a deleterious mutation in BRCA1/2 was not confirmed in 22 cases. Twelve of these discrepancies were due to differences in variant classification (i.e., different criteria between F1-CDx and local testing assays); the remaining 10 patients actually had LR (≥1 exon deletions or duplications), indicating a substantial lack of sensitivity for LR detection from the F1-CDx assay; however, this resulted in FDA approval of F1-CDx as a CDx for this indication [129]. Recently, FoundationMedicine has marketed the FoundationOne Liquid® CDx for detecting tBRCA* (with a parallel analysis of the 324-gene panel) directly in a blood sample; however, this test does not detect LOH and has only been validated prior to rucaparib treatment [130]. It should be noted that multigene panel testing, although time- and cost-efficient due to its comprehensive content, also increases the risk of detecting VUS. Interestingly, although FF-CDx for tBRCA* detection is a prerequisite for PARPi prescription for two distinct indications in OC, the third FDA-approved indication (HRD evaluation prior to rucaparib maintenance as a second-line treatment) is not biomarker-driven. Indeed, a positive HRD status is considered predictive of efficacy and to indicate enhanced PFS.The development of MC-CDx, which is supported by MG’s expertise with BRACAnalysis®, has also advanced; its GIS calculation differs from that of FF-CDx. The correlation between LOH, LST and TAI with BRCA1/2* and Pt-S has been previously described [80,81,82]; subsequently, the superiority of the correlation among these three measurements was shown in comparison to each individual component [131]. Consequently, the GIS (proprietary score of MG- GISMG) consists of the unweighted numeric sum of LOH, LST and TAI. HRD positivity is currently defined by a GISMG ≥ 42 and/or tBRCA1/2*. The GISMG ≥ 42 threshold was set following the analysis of a training cohort that consisted of 497 chemotherapy-naïve BC patients and 561 EOC patients whose BRCA1/2 status was known. Thus, by evaluating HRD scores in this cohort, the GISMG positivity cutoff was predefined with a 95% sensitivity for detecting tumors with BRCA1/2* or BRCA-CpG+; this test was then evaluated for its ability to identify Pt-S triple-negative BC (TNBC) in a neoadjuvant setting [132].Following its initial validation in TNBC, MC-CDx was investigated in several RCTs, such as PRIMA, VELIA and PAOLA-1, which evaluated PARPis (niraparib, veliparib and olaparib, respectively) in EOC [133,134,135]. Importantly, similar to FF-CDx, the GIS positivity threshold of MC-CDx (GISMC) has been set at different values. Indeed, while it was initially proposed at GISMG ≥ 42, another cutoff of GISMC ≥ 33 (corresponding to the first percentile of HRD scores observed in tBRCA1/2* tumors) has also been used, such as in the VELIA trial, with the goal of avoiding false-negatives [42,46,134].Similar to FF-CDx, MC-CDx relies on a multistep process based on hybridization-based capture and NGS (i.e., fragmentation, end repair and adenylation, adapter ligation, library construction/amplification, hybridization and capture, sequencing and data analysis). The hybridization process was performed through a custom Agilent SureSelect© capture array consisting of pangenomic probes at 54091 SNV sites and 685 probes for BRCA1/2 exons and exon boundaries. Normalized base and exon coverage of BRCA1/2 were calculated to detect LRs. MC-CDx exhibited a less sensitive performance in BRCA1/2* detection (notably for LRs) than BA-CDx, as it is performed through NGS and on biopsies (implying tissue heterogeneity). Indeed, indels >25 bp were less frequently detected than whole gene duplications/deletions. The LODs were as follows: 7.2% for an SNV, 6.6% for a <10 bp deletion, 6.3% for a <10 bp insertion, 5.9% for a ≥10 bp deletion, 30% for ≥3 exons LR and 50% for 1–2 exons LR.For BA-CDx, MC-CDx uses the MG proprietary classification score for variant classification. Comparison with a validated NGS-based assay on 209 FFPE clinical specimens from cancer patients (5 tBRCA1/2* GISMC −; 71 tBRCA1/2WT GISMC −; 66 tBRCA1/2* GISMC +; 61 tBRCA1/2WT GISMC +) indicated 99.9% PPA (95% lower limit confidence of 99.7%), 100% NPA for BRCA1/2 SNVs/indels and 100% OPA for tBRCA LR. Concordance analysis gave high fidelity results, both for GIS status (98.5% OPA, 97.4% NPA and 98.1% OPA) and HRD status (98.5% OPA, 98.6% NPA and 98.5% OPA). In 136 FFPE samples harboring tBRCA1/2WT, a 0% false-positive rate was observed, and a MAF threshold of 5% was defined, as no spurious variants were observed above this cutoff. The MC-CDx results were not affected by necrosis of the tumor area up to 60%.The SOLO1 sample (n = 391), already tested with BA-CDx (in whole blood) for gBRCA*, retrospectively provided clinical validation of MC-CDx. FFPE DNA samples from 298 patients were used: 284 patients were confirmed to carry tBCRA*, 8 were not, and 6 samples failed the test. Pathogenic LR, which was detected at the germline level by BA-CDx in 15 patients produced the following results within the tumors: detection in 12 (80%) cases, absence of detection in 1 (6.7%) case because of the known limit of MC-CDx detection and 2 (13.3%) unanalyzable samples because of the poor quality of the tumor specimens. In the almost “real-life” conditions of clinical trials, MC-CDx indicated an unknown HRD status in 10–20% of patients: 10% (46/463 patients-all causes) in QUADRA (NCT02354586), 12% (137/1140 patients-all causes) in VELIA (NCT02470585), 15% (54/350; 26 inconclusive results and 28 inadequate/missing specimens) in NOVA (NCT01847274), 20% (163/806; 70 inconclusive results and 93 inadequate/insufficient specimens) in PAOLA-1 (NCT02477644) and 15% (111/733; 80 inconclusive results and 31 inadequate/insufficient specimens) in PRIMA (NCT02655016). Notably, when focusing on trials that detailed the cause of the lack of HRD information (i.e., NOVA, PAOLA-1 and PRIMA) and by examining only the missing data caused by failed tests (i.e., inconclusive results in tested patients), it appears that MC-CDx fails to provide a valid result in 8.1–11.4% of tests.As a consequence of this prospective validation, which was based on RCTs and will be detailed in the companion paper (Part 2), these two FDA-approved assays led to a better definition of therapeutic sensitivity (notably for PARPis) and have been included as part of the EOC management in the European Society for Medical Oncology (ESMO) and American Society of Clinical Oncology (ASCO) recommendations [11,115].Despite yielding considerable progress in HGSOC management (which is detailed in the companion paper), validated HRD assays currently suffer from several technical limitations that can be schematically categorized as preanalytical, analytical or postanalytical. Furthermore, medical considerations, which will be developed in the companion paper, should also be assessed.The technical concerns can be further divided into three categories: preanalytical, analytical and postanalytical. First, the different FDA-validated HRD assays (i.e., BA-CDx, MC-CDx and F1-CDx) do not exhibit the same performance on clinical samples and are therefore not interchangeable. When comparing MC-CDx (positive if ≥42%), the percentage of LOH (analogous to the measurement of HRD status through F1-CDx; positive if ≥16%) and an 11-gene panel (consisting of genes involved in the HR pathway; positive if a pathogenic variant is found), it appears that many HRD+ patients (according to MC-CDx) are not detected through the other methods. Indeed, up to 46% of HRD+ (MC-CDx) patients were missed by the percentage of LOH and the 11-gene panel. Moreover, using a lower cutoff (i.e., GIS ≥ 33%), such as used in the VELIA trial, resulted in missing up to 61% of patients [136].According to the different RCTs, approximately 5–10% of samples are unfit for HRD processing. This can be the consequence of insufficient starting material, paucity of tumor cellularity, or poor quality or degraded specimens.Another aspect that should be considered is sample heterogeneity, which occurs at two distinct levels. Typically, a processed sample exhibits < 100% tumor cellularity, meaning that the analysis will include a fraction of normal tissue that can interfere with HRD evaluation, either by lowering the GIS or preventing tBRCA* detection; importantly, this limit does not exist for gBRCA*, as all cells harbor the mutation. Moreover, each tumor exhibits different lineages and clonal evolutions, with possible discordant HRD statuses. In recent years, tumor heterogeneity has emerged as a cornerstone of treatment resistance, including resistance to platinum and PARPis. This intratumoral heterogeneity is also present between distinct tumors, for instance, between primary and secondary lesions [137].BA-CDx, while being the gold standard for gBRCA analysis, detects only germinal mutations and consequently fails to detect sBRCA*, which occurs in approximately 30% of BRCA-mutated HGSOC cases. Therefore, a negative result from the BA-CDx blood test does not rule out the presence of sBRCA*.Setting aside BA-CDx, the tumoral HRD assays suffer from high LODs (explained in the companion review). While manufacturers indicate that a minimum value of 20% tumor cellularity is required for processing, at least 30–35% tumor cellularity is needed to evaluate GIS (or risk a false-negative). Furthermore, beyond the LODs related to tumor cellularity, HRD assays evaluating tBRCA* barely detect LRs, although LRs are present in up to 16% of HGSOC patients, thus leading to false tBRCAwt statuses [33]. Recently, SIGNPOST showed that approximately 20% of gBRCA* was missed by the initial tBRCA* assessment, with the stunning revelation that none of the LRs of the cohort (representing 11% of gBRCA*) were detected by the tBRCA* evaluation [138].Even if the starting material is sufficient for the requirements of the manufacturer, approximately 5–10% of processed samples still give inconclusive results regarding the HRD status. For BRCA1/2 genes, the main concern is misclassification. Although recent years have led to an improved understanding of the functions and molecular alterations of BRCA1/2 (and consequently to more accurate classifications), discrepancies still exist within databases [139,140,141]. While the accumulation of data led to important refinements in the classification of BRCA mutations, and despite the synchronization of classification according to guidelines and data sharing, reclassification still frequently occurs [142]. In addition to clear pathogenic variants, “likely pathogenic” variants are frequently included in RCTs and in FDA/EMA approval, although the latter does not result in absolute proof of its pathogenic nature. Importantly, some variants are considered “pathogenic” based on a proven deleterious effect observed in a germline context. This “pathogenic” status is then extrapolated to sBRCA*; however, gBRCA* and sBRCA* may have distinct effects at both the molecular and cellular levels, just as BRCA1 and BRCA2 do not share exactly the same functions. Furthermore, tBRCA analysis does not define zygosity; indeed, tBRCA* can be homozygous and/or heterozygous in the sample, potentially leading to distinct effects. Orthogonal to clinical considerations, F1-CDx (which analyses over 300 genes unrelated to HR and other DNA alterations) also assesses “off-target” multigene testing, raising the risk of detecting VUS without increasing the possibility of detecting pathogenic variants (and subsequently proposing targeted therapy) [143].At the GIS level, as discussed in the previous section and in the companion paper, defining the ideal threshold value is a matter of heated debate. Indeed, beyond providing the most accurate and precise HRD assay, the main risk of using an unfit threshold is misclassification. This falls into two categories: false-positives (FP) and false-negatives (FN). An FP represents an HRD-positive result when the sample was actually HRD-negative (or HR-proficient); an FN is when an HRD-positive sample is labeled HRD-negative. In MC-CDx, the test is positive if the GISMG is ≥42 (initially validated with TNBC and corresponding to a 95% sensitivity for detecting tumors with BRCA alterations); this test has been evaluated with different cutoffs. For instance, in the VELIA trial, a lower cutoff (GIS ≥ 33) was used, with the aim of preventing FNs. Moreover, within BRCAwt patients, HRD is not a predictive biomarker for PARPi sensitivity. Conversely, a retrospective study based on TCGA data and using a backward strategy (i.e., moving from GIS-based stratification toward clinical/molecular data) showed that using a threshold of GIS ≥ 63 led to accurate HGSOC classification among the subtypes, correlating well with prognosis and the presence of BRCA1/2* [144]. Consequently, the “magic 42” still needs to be more precisely defined, with potential variation according to tumor type and clinical context.Aside from these technical issues, medical considerations should also be taken into account, particularly the timing of HRD evaluation, to obtain a broader perspective. These considerations will be developed in Part 2 of this review.Emerging strategies for HRD assessment mainly occur along three axes: molecular tools for HRD assessment, dynamic assays (i.e., functional assays) for evaluating HRD status, and more global strategies (including nomograms). One of the main considerations for refining HRD evaluation, beyond its relevance, is that a biomarker described in basic research should be feasible in clinical practice, taking technical, economic and temporal issues into account. For instance, metabolomics studies or spheroid cultures have shown that HRD tumors exhibit a specific profile; nevertheless, to date, these techniques appear unfit for routine clinical application [100,145].Several strategies regarding molecular assays to determine HRD have been deployed, with various results. Notably, the primary aim is improved identification of patients who would benefit from frontline treatment with PARPis, as these drugs suffer the same limitations as existing CDx (i.e., genomic scars). Many private companies have developed their own tools, but they will not be discussed here, as those tools failed to show better results than those of MC-CDx. However, some interesting tools will be outlined. Signature 3, which is based on SNVs, correlates with HRD and platinum sensitivity [71]. A relevant tool, entitled Signature Multivariate Analysis (SigMA), allows HRD-associated mutational signatures to be detected directly from targeted gene panels [146]. HRDetect, which is based on whole-genome sequencing and the incorporation of six distinct HRD-related signatures that predict BRCA1/2 alterations, is a promising assay that provides almost 100% detection sensitivity [147]. Both signature 3 and HRDetect are “backward strategies”, meaning that they predict a BRCA1/2 status from secondary molecular signatures. A recent paper showed that HRDetect outperforms %LOH (F1-CDx) and is equally efficient as GISMC for HRD identification [148]. Another emerging field is represented by the analysis of LRs; biomarkers, such as CCNE1 and ESMY amplifications within LRs are associated with HR proficiency and deficiency, respectively [9,45,149]. Furthermore, specific profiles of LRs are associated with better prognosis and with Pt-S, implying potent sensitivity to PARPis [150,151]. Other promising candidates, such as RAD50 deletion, RB1 loss and BRD4 amplification, could improve HRD detection [43,152,153].While mutations in non-BRCA HRR-related genes provide conflicting results (only homozygous deletion in PTEN or CHK1 is putatively associated with HRD), a more accurate evaluation could come from integrating clonal composition [144]. Indeed, by integrating NGS metrics (i.e., determining if variants correspond to clonal or subclonal mutations), it was recently shown that mutations in HRR-related genes were associated with OS and Pt-S only if they were clonal [154].At the epigenetic level, BRCA1 (and, to a lesser extent, BRCA2) methylation profiles should not be overlooked, although they are not routinely examined. The frequency of methylation is not anecdotal, as it has been reported in 19.3% (BRCA1) and 4.6% (BRCA2) of 92 BRCAwt cases, according to a retrospective study. Acquired loss of RAD51C promoter methylation is associated with PARPi resistance: more precisely, the presence of heterozygous methylation leads to resistance, while homozygous methylation leads to sensitivity [155]. Several studies have assessed HGSOC methylomes and revealed distinct methylation profiles linked to Pt-S or, conversely, primary/secondary platinum resistance (Pt-R); these could constitute putative biomarkers to evaluate in the context of PARPis [156,157,158].Beyond deciphering specific alterations, such as DNA mutations or epigenetic markers, comprehensive approaches that collect and assemble common genomic, epigenomic and functional data would result in better molecular dissection and the development of new biomarkers [159,160]. Indeed, by developing multilayer (i.e., genome/exome, SVs, transcriptome, miRNome, proteome, methylome and proteome) and integrated molecular identity cards (“multiomics”), such as that performed by the TCGA but with a more HRD-oriented view, we could subsequently select “core biomarkers” with higher sensitivities and specificities, leading to a more accurate evaluation of HRD status [161,162,163,164]. As such, rather than extensive and expensive approaches that would be difficult to translate to clinical practice, we could define a subset of distinct biomarkers through distinct techniques that would increase the sensitivity/specificity of current CDx assays and decrease the number of inconclusive cases.Sequencing ctDNA in blood samples, which has already been validated by the F1 L-CDx assay, allows direct evaluation of a tBRCA1/2 status [165]. In the near future, an iterative analysis could provide clues as to primitive or secondary resistance, depending on the presence of reverse mutations, as it has been shown that they are associated with resistance to PARPis [166]. One method would be via direct assessment of GIS in blood samples. Although it is not currently performed on ctDNA, intriguing papers have shown that it is feasible to analyze SVs in blood [167,168,169]. The molecular signatures of miRNAs present in blood samples have been detected and showed an association with early OC and prognosis, thus demonstrating their feasibility as potent surrogate markers for HRD [170,171,172,173]. Furthermore, other serum biomarkers are under study and currently debated, but their translation into clinical practice seems difficult [174]. Moreover, ascitic cancer cells should be considered, as they provide a means to obtain material with less invasive processes than biopsies; these cells are easily obtained and well represent the mutations (including SVs), DNA methylation and intratumoral heterogeneity found in EOC [175]. Thus, “liquid biopsies” could represent an easy-to-use method for iterative sampling [176].At the functional level, several techniques have been developed in the past decade with the aim of evaluating the actual HRD status [99]. Indeed, apart from investigating the causes (e.g., DNA mutations) and consequences (e.g., genomic scars) of HRD, functional assays directly monitor the process itself. Functional tests can potentially overcome the inherent limits of current assays, which require either a genetic alteration (i.e., affecting non-BRCA HR-related genes) or the calculation of GIS (which reflects a history of HRD). However, similar to current CDx assays, they do not predict the sensitivity of HRD-unrelated PARPis.One of the techniques with a high potential for clinical applications is the REcombination CAPacity (RECAP) assay [177]. When a DSB occurs and is managed with HR, RAD51 (one of the downstream effectors of BRCA1/2) attaches to these sites to facilitate sister chromatid invasion. Thus, measuring RAD51 nuclear foci provides a direct evaluation of the efficiency of HR, irrespective of its etiology. In the context of HRD, these foci are absent.HRD evaluation through RAD51 foci was initially developed in the laboratory and subsequently applied in different protocols, as ex vivo or through patient-derived xenograft assays, to predict patient sensitivity to platinum/PARPis and OS [178,179,180,181]. One of the main drawbacks of the RECAP assay is that it requires fresh tissue as the starting material as well as the induction of DNA damage, since the test relies on tissue irradiation and subsequent visualization of foci formation (with distinct cutoffs of foci corresponding to functional or deficient HR), limiting its application in clinics (as tumoral tissue is frequently processed as FFPE). However, as HRD tumor cells exhibit spontaneous important DSBs, an evolution of these assays has emerged. Though originally focused on BC samples, the assay was subsequently directly performed on FFPE samples, showing a correlation between RAD51 foci and HRD status [182,183]. Recently, an adaptation to ovarian and endometrial cancer FFPE tissues, entitled RAD51-FFPE, was developed [184]. Interestingly, RAD51-FFPE paved the way for clinical applications by optimizing the test via calibration of a threshold corresponding to functional HRD. This led to high sensitivity, as it allowed BRCA-deficient and HRD tumors to be detected in 90% and 87% of cases, respectively. Therefore, the next step will be its integration into clinical studies to evaluate its performance as a predictive biomarker.HGSOC, the most frequent and aggressive form of OC, represents an important challenge for researchers and clinicians. Half of these cases show HRD, which has specific causes and consequences. In terms of etiology, HRD is mainly caused by genetic and epigenetic alterations, with BRCA1 and BRCA2 best characterized thus far. In addition to BRCA1/2, many other lesions are involved in the etiology of HRD, leading to the development of the BRCAness phenotype. HRD has specific consequences at both molecular (e.g., genomic instability) and clinical (e.g., PARP inhibitor sensitivity) levels.Based on its prevalence and its theranostics potency, HRD represents a major molecular factor that must be understood to improve HGSOC management. Three CDx assays currently have FDA approval for the identification of HRD status, helping physicians prescribe PARP inhibitors. However, there is an urgent need for novel assays, such as functional assays, to be developed and integrated into RCTs for clinical validation.S.Q. wrote this review with inputs from M.F. and J.S. All authors have read and agreed to the published version of the manuscript.Montpellier University Hospital.The authors declare no conflict of interest.Genomic instability as a consequence of HRD. HRD shows genomic instability and pro-oncogenic lesions responsible for multiple genetic (blue squares) and epigenetic (green squares) events characterizing tumor progression and defining PARPi sensitivity and survival enhancement (black squares). Signature 3 refers to a specific base substitution pattern, which is the consequence of HRD. Abbreviations are as follows: HRD: homologous recombination deficiency; LOH: loss of heterozygosity; LST: large-scale transition; MMEJ: microhomology-mediated end-joining; OS: overall survival; PFS: progression-free survival; PARPi: poly (adenosine diphosphate-ribose) polymerase inhibitor; PTM: posttranslational modification; TAI: telomere allelic imbalance.Current FDA-approved CDx for HRD evaluation in ovarian cancer.Abbreviations are as follows: 1Lm = first-line maintenance; 2Lm = second-line maintenance; ≥2 L = 2 or more previous lines of chemotherapy; ≥3 L = 3 or more previous lines of chemotherapy; aEOC = advanced epithelial ovarian cancer (including primitive peritoneal and fallopian tube cancers); CDx = companion diagnostic; C/PR = complete/partial response to platinum-based chemotherapy; FFPE = formalin-fixed paraffin-embedded; FM = FoundationMedicine; gBRCA = germline BRCA; GIS = genomic instability score; HD: homozygous deletion; HRD (+) = homologous recombination deficiency (positive); LOH = loss of heterozygosity; MG = MyriadGenetics; NA = not applicable; NIRA: niraparib; OC = ovarian cancer; OLA: olaparib; Pt-s = platinum-sensitive; rEOC = recurrent epithelial ovarian cancer (including primitive peritoneal and fallopian tube cancers); RUCA: rucaparib; tBRCA = tumoral BRCA. Nota bene: 1 data is presented according to the technical information provided by the manufacturers and laboratory validation studies; complementary technical information, particularly for clinical validation studies, is provided within the text. 2 The cost of BRACAnalysis® CDx is highly dependent on medical insurance coverage in the USA. 3 The F1-CDx also detects microlesions and macrolesions in 324 genes, selected gene rearrangements, and MSI and TMB. 4 In this context, FoundationOne Liquid CDx (performed on a whole blood sample) is also FDA-approved but does not provide LOH evaluation. 5 In this context, rucaparib is not biomarker-driven, but a positive HRD status is predictive of its efficacy and indicates improved progression-free survival.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01133.txt

@@ -0,0 +1 @@
+Salivary gland cancer is rare and there is a need to develop new and effective drug therapies. New drugs are in development in the recurrent and metastatic setting that target specific changes within a cancer, some of which can be detected through sequencing of the cancer DNA. This study addressed how useful the DNA sequencing of cancer samples is to inform the decision of which drug therapy within a trial is the best match for each individual patient. We found that using focused DNA panels, including small numbers of genes, helped to match just over one in four patients with salivary gland cancer to drug therapies. The matching rate of the focused panel varied by subtype and was least useful in adenoid cystic carcinoma (ACC), at 7%. However, in ACC, larger gene panels had added value, identifying matched trial therapies in 40% of cases.For most patients with salivary gland cancer, there are no effective standard systemic therapies. Although clinical trials of biomarker-led drug therapies have delivered significant recent advances, there remains a need to understand the clinical utility of genomic profiling of cancer as a means to match patients with recurrent or metastatic salivary gland cancer to clinical trial therapies. In total, 209 patients with salivary gland cancers were profiled with 24 gene (n = 209)) and >325 gene (n = 32) DNA-based next-generation sequencing panels. A retrospective systematic evaluation was performed to identify the frequency of available matched drug therapies within clinical trials based on the results. The matches were then stratified based upon the level of evidence supporting the drug–biomarker combination being investigated using the ESMO Scale for Clinical Actionability of Molecular Targets (ESCAT) to determine the strength of the clinical rationale for each gene–drug match identified. DNA-based next generation sequencing (NGS) analysis was successful in 175/209 (84%) patients with salivary gland cancer. Using the 24-gene NGS panel, actionable alterations were identified in 27% (48/175) patients. Alterations were most frequent in salivary duct carcinoma (88%) characterized by TP53 and/or PIK3CA mutations, with matched trials available for 63% (10/16). In ACC, biomarker-matched trials were available for 7% (8/115), and no genomic alterations were found in 96/115 (83%) of ACC patients. TP53 was the most frequently altered gene across all subtypes; however, there were no trials recruiting based on TP53 status. In 32 ACC patients with no genomic alterations using the 24-gene panel, a broader (>325 gene) panel identified alterations in 87% (27/32) of cases with biomarker-matched trials available in 40% (13/32) cases. This study identified that genomic profiling using focused (24-gene) NGS panels has potential utility in matching to trial therapies for most patients with non-ACC salivary gland cancer. For patients with ACC, broader genomic profiling has demonstrated added clinical utility. We describe the application of an approach to classification of levels of evidence which may be helpful to inform the clinician and patient decision making around the selection of clinical trial therapies.Salivary gland cancer (SGC) is a rare disease comprising over 24 histopathological subtypes [1]. Although many patients are cured following surgical resection with or without adjuvant radiation, consensus guidance on the optimum management of recurrent or metastatic disease is based on limited evidence and hindered by the relative paucity of clinical trials incorporating these patients [2].Although some patients may derive meaningful benefit from cytotoxic chemotherapy [3], there remains limited trial evidence to support its routine use. The most significant recent advances have come from clinical trials of biomarker-directed drug therapies [4,5,6,7,8,9,10,11,12]. For example, the amplification of Erb-B2 receptor tyrosine kinase 2 (ERBB2/HER2) protein overexpression is seen in 20–30% of cases of salivary duct carcinoma [13] and there is trial data to support the efficacy of HER2-targeting therapy with either trastuzumab and docetaxel [11], trastuzumab and pertuzumab [12] or trastuzumab-emtansine [10] in recurrent or metastatic ERBB2-amplified/HER2-overexpressing disease. In addition to HER2, the androgen receptor (AR) is overexpressed in most patients with salivary duct carcinoma (SDC) [13]. There are clinical trial data showing the efficacy of either androgen deprivation therapy with bicalutamide [13] and enzalutamide [9] and combined androgen blockade with bicalutamide and triptorelin [8] in recurrent or metastatic AR-overexpressing salivary duct carcinoma.The other subtype of salivary gland cancer which has been most impacted by the development of biomarker-led therapy is secretory carcinoma. Chromosomal rearrangements resulting in fusions of the Neurotrophic Tyrosine Receptor Kinase (NTRK) gene are described in around 90% of cases of secretory carcinoma [14] and there are data showing the efficacy of entrectinib [4] and larotrectinib [5] in NTRK-rearranged tumours from any primary site, including NTRK-rearranged salivary gland cancers. Previous studies in SGC identified TP53 and the PI3K pathway as the most commonly altered through point mutations across SGC subtypes. Mutations in ERBB2, EGFR, and BRAF have previously been detected at low frequencies and have the potential to match patients to targeted therapies [15].Aside from these few examples of biomarker-directed therapies relevant to salivary gland cancers, in order for clinicians to engage in informed discussion with their patients on the potential role for genomic profiling, there is a need to better understand the clinical utility of these approaches. In addition, clinicians increasingly encounter situations in which patients attend consultations seeking an interpretation of genetic sequencing data on their own tumours that they have obtained from commercial vendors. We, therefore, sought to evaluate the frequency with which DNA-based next-generation sequencing (NGS) yields additional information to aid selection of clinical trial drug therapies for a cohort of patients with salivary gland cancer and to judge the strength of the underlying rationale for individual biomarker–drug matches in this population.We performed a retrospective cohort study on 209 patients with salivary gland cancers who underwent clinical review at a tertiary cancer centre (The Christie NHS Foundation Trust, Manchester, UK) from April 2017 to December 2020. Patients provided informed consent for the collection of genomic, clinical and demographic characteristics. This study was granted research ethics approval under the Manchester Cancer Research Centre Biobank Research Tissue Bank Ethics (NHS NW Research Ethics Committee 18/NW/0092) and was performed in accordance with the Declaration of Helsinki.For all patients, a focused DNA-based NGS panel was performed including 24 cancer-related genes through the course of this study. The genes included and regions of each gene covered are shown in Supplementary Table S1. DNA was extracted from archival FFPE samples, and samples were requested to have a minimum tumour content of 20% and analysed using a Qiagen GeneRead DNAseq Targeted Panel V2 in the National Health Service Northwest Genomics Laboratory Hub. A median exon coverage depth of >350× was performed and a customised bioinformatic pipeline was validated to detect single nucleotide variants and indels (<40 bp) to 4% mutant allele frequency. Where identified, variants were classified using Cancer Variant Interpretation Guidelines UK [16] and tiering from the joint Consensus Recommendation of the Association for Molecular Pathology, American Society of Clinical Oncology, and College of American Pathologists [17] with reference to publicly available resources including Catalogue Of Somatic Mutations In Cancer v19, and other subscription-based resources including Human Gene Mutation Database Professional (Qiagen). In a secondary analysis, on samples from 32 patients with adenoid cystic carcinoma, in whom no alterations were identified using the focused panel, repeat analysis was performed using a broader DNA-based NGS panel. For these samples, the commercial Foundation Medicine assay was performed in a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory and accredited by the College of American Pathologists. Tumour DNA underwent hybrid capture for the entire coding region and select introns from genes frequently mutated or rearranged in cancer (including >325 selected cancer-related genes through the course of this study). Illumina HiSeq NGS was performed with a median exon coverage depth of >500× to identify single-nucleotide variants, indels, copy number alterations and select gene rearrangements (as described in [18]). Summary findings of NGS data were visualised on cBioPortal [19,20].To link genetic variants to drug therapies within clinical trials, the 24 genes included in the focused panel were queried in the public ally available databases OncoKB [21] and MyCancerGenome.org to identify pre-existing biomarker–therapy matches and biomarker-led studies, respectively. To identify preclinical evidence for biomarkers associated with response to targeted therapies, literature searches of PubMed were undertaken using the search terms “GENE NAME” and (“predictive biomarker” OR “Targeted therapy” OR “Response” OR “sensiti*”). To determine whether any clinical trials were available in the UK incorporating these therapies and open to patients with salivary gland cancers, a search was performed on the clinicaltrials.gov database for trials open to recruitment to patients with salivary gland cancers or any solid tumour type using therapies established from the above process with sites active in the UK from 2016 to December 2020.Genomic findings were ranked using the European Society of Clinical Oncology (ESMO) Scale for Clinical Actionability of Molecular Targets (ESCAT) scoring system [22] independently by two co-authors (S.R. and L.W.). When different scores were allocated, a further review was undertaken by R.M. and a final consensus reached. The ESCAT scaling defines six tiers of clinical evidence supporting the actionability of genomic–drug matches as follows: Level I, genomic–drug matches ready for implementation in routine clinical decisions; level II, genomic–drug matches as investigational targets that are likely to define a patient population that benefits from a targeted drug but additional data are needed; level III, genomic–drug matches with clinical benefit previously demonstrated in other tumour types or for similar molecular targets; level IV, genomic–drug matches with preclinical evidence of actionability; level V, genomic–drug matches with evidence supporting co-targeting approaches; and level X, genomic findings with lack of evidence for actionability. For the ESCAT classification of each gene under investigation, the highest score was taken when different specific variants within the same gene carried different scores.A total of 209 patients with salivary gland cancer reviewed at a single tertiary cancer centre between 2017 and 2019 were included in this study. The clinical characteristics of the patients are shown in Table 1. Consistent with a population of patients seeking clinical trial therapies, the median age was 51 years (range: 23–78 years) and almost all patients (91%) had radiological confirmation of recurrent or metastatic disease with a measurable component at the time of review. In total, 44% of patients had tumours arising in the minor salivary gland consistent with the relatively high frequency of disease recurrence in this population [23]. ACC was the most frequent histopathology (68%), as would be expected given the relatively high risk of recurrence in this subtype. Of the remaining patients with non-ACC SGC, adenocarcinoma not otherwise specified (NOS) and SDC were most frequent.To facilitate personalised clinical trial selection, a focused DNA-based NGS panel including 24 genes frequently associated with somatic mutations in cancer (Table S1) was applied to FFPE tumour samples from 209 patients with salivary gland cancers being reviewed to consider clinical trial therapies. To determine the utility of this panel to guide clinical trial selection, we performed a systematic evaluation of genomic biomarker–drug trial matches. Table 2 summarises the genes included in this panel and the matched drug therapies available through the course of this study.Although all of the genes included in this panel had the potential to match patients to drug therapies, through the duration of this study, matched drug therapies within trials were open to SGC/solid malignancy patients for 63% (15/24) of the sequenced genes. To classify the level of evidence supporting the genomic biomarker under investigation, genomic findings were ranked using the ESCAT scoring system [22]. None of the genomic alterations was classed as level 1, defined as the drug–target match being associated with improved outcome in clinical trials in the specific tumour type. The highest ESCAT score was for PIK3CA at level 2b, as clinical trial results of drugs targeting this pathway have demonstrated a signal of increased radiological response rate without data showing a meaningful overall survival benefit [38].The remainder were classed as level 3 (12/15), defined as demonstrating clinical benefit in other tumour types or in similar variants, or level 4a (2/15), defined as actionability of target predicted in preclinical in vitro or in vivo models.Applying the 24-gene NGS panel to a consecutive series of 209 patients with salivary gland cancer considering trial therapies, the analysis through NGS was successful in 175 cases (84%). The remainder failed due to insufficient or poorly preserved DNA. Figure 1 shows the results of the focused NGS analysis on 175 patients with salivary gland cancer. Sixty-three variants of significance were identified in 48 patients (27%). The most frequently altered genes were PIK3CA and TP53, which made up 61% (39/63) of all alterations identified. The frequency of alterations was lowest for patients with ACC (19/115, 17%) compared with non-ACC SGC (29/60, 50%). The highest frequency of alterations seen using this approach was for patients with SDC which was characterised by the presence of TP53 and/or PIK3CA mutations in 14/16 cases. Patients with myoepithelial, neuroendocrine (n = 1), NUT (n = 1) and secretory (n = 1) carcinomas had no identifiable mutations with this panel.TP53 alterations were the most frequent finding, with 34 mutations identified in 33 samples (Table 3). These were missense mutations in 19/34, and 12/34 were truncation mutations. One sample had a silent mutation, TP53 c.375G > A p.(Thr125Thr). However, this variant has been shown to lead to in intron inclusion between exons 4 and 5 and is therefore pathogenic. Depicted as “other” in Figure 1, further mutations occurred at splice sites, resulting in pathogenic changes. Although TP53 alterations were the most frequent finding, the drug–biomarker combinations are under investigation in trials based upon pre-clinical (level 4a evidence), and there were no open trials with a rationale for the inclusion of patients with TP53 mutations through the course of this study. Level 4a was allocated based on preclinical data with ATR inhibitors. TP53-mutant tumours were shown to have increased sensitivity to ATR inhibitors in combination with chemotherapy or radiotherapy [42].In total, 50% (8/16) of patients with SDC had mutations in PIK3CA/AKT/PTEN, conferring the hyperactivation of the PI3K signalling pathway. In ACC, the frequency was 5% (6/115). In PIK3CA, 12 gain-of-function variants were detected in 11 patients (Table 3). As such, 2/115 ACC patients (2%) could have been matched to trials based on their PIK3CA mutation status with level 2b evidence. Level 2b was attributed to PIK3CA mutations based on preliminary data published from basket trials including PI3K inhibitors, showing a response rate of 16% and stable disease rate of 66% at 6 months in patients with tumours harbouring PIK3CA mutations [38]. The utility of PIK3CA rises to 14% (9/63) when applied to nonadenoid cystic salivary gland cancer and to 37% (6/16) for patients with salivary duct carcinoma. However, in non-ACC SGC, the evidence level is 3a based on the efficacy of PIK3CA inhibitors in breast cancer [43]. One case of apelisib plus androgen deprivation in SDC is reported, resulting in significant benefits [44].AKT1 gain-of-function (E17K) variants were identified in 1% (1/115) of ACC and 13% (2/16) of SDC patients. AKT1 inhibitors have shown higher response rates in a basket trial in patients with AKT1 E17K-mutated tumours; however, no SGC patients were included in this trial and they are, therefore, ranked as 3a.Loss-of-function PTEN variants were observed in 3% (4/115) of ACC patients and 3% (2/63) of non-ACC SGC patients. Clinical responses have been seen in the early phase trials of PI3Kb inhibitors in tumours with PTEN loss in prostate cancer [39] and gastric cancer [45], giving PTEN a 3a ranking.Pathogenic ERBB2 mutations were identified in three patients with salivary gland cancer (Table 3), which provides a match to trial therapies with level 3a/3b evidence. Level 3b is applied to SDC as there is evidence of efficacy of anti-HER2 therapy in HER2-amplified SDCs [13] and in other SGC subtypes. It was classified as 3a as a result of the efficacy of anti-HER2 therapies in lung cancer patients with gain-of-function mutations [46]. Again, this alteration was most frequent in patients with salivary duct carcinoma (2/16, 13%).EGFR was mutated in one patient with adenoid cystic carcinoma (Table 3). The variant was c.2369C > T p.(Thr790Met), which is a commonly acquired variant that confers resistance to most EGFR tyrosine kinase inhibitors, but is sensitive to Osimertinib [30]. This matched these patients to trial therapies with level 3a evidence due to the efficacy data for non-small-cell lung cancer.Two BRAF alterations were identified in patients with adenoid cystic carcinoma and salivary duct carcinoma (Table S2). Both were class 3 gain-of-function mutations. Class 3 BRAF mutations are characterized by a dependence on RAS signalling and are hypothesized to be sensitive to RAS inhibition. While BRAF V600E would be classified as 3a, BRAF class 3 mutations are classed as X as they have no kinase activity themselves and are dependant on upstream oncogenic signalling and commonly co-occur with NF1 deletions [47]. A single KRAS mutation was identified in a patient with an adenocarcinoma; this missense mutation resulted in KRAS gain of function. The variant identified was not G12C and, as such, there is only preclinical evidence of increased sensitivity to MEK/ERK inhibitors. Thus, this variant was classified as 4a [36].CTNNB1 codes for beta-catenin: One variant in an ACC patient was detected. Preclinical data have shown CTNNB1 variants to confer increased sensitivity to CBP/Betacatenin inhibitors, and so this variant was therefore classed as 4a [28].Through the course of this study, using the 24-gene panel, 7% (8/115) of ACC patients could be matched to a biomarker-led clinical trial based on the molecular screening results. In 6% (7/115) of cases, this was due to variants conferring the hyperactivation of the PI3K pathway (AKT1, PIK3CA and PTEN variants), who could be treated with AKT1 inhibitors, such as Capivasertib (NCT0122631, NCT02338622). The remaining ACC match was for a patient with an EGFR mutation, who could have been matched to AFM24 (NCT04259450).In SDC, 63% (10/16) could be matched to biomarker-led trials. A total of 50% (8/16) were, again, matched to AKT1 inhibitors such as Capivasertib and 13% (2/16) to HER-2 inhibitors such as Neratinib (NCT01953926) or TAS0728 (NCT03410927).Matched trial therapies were available for 20% (3/15) of patients with adenocarcinoma (NOS), 13% (2/15) to AKT1 inhibitors (NCT01226316, NCT02338622) and 7% (1/15) to RAF/MEK inhibitors, based on a KRAS mutation through NCT02407509. The relatively small number of patients included with other non-ACC subtypes limits the analysis. However, matched trial therapies were available for 25% (2/8) of patients with carcinoma ex-pleomorphic adenoma, 20% (1/8) with acinic cell carcinoma, and 16% (1/6) with muco-epidermoid carcinoma.As 81% of ACC patients (94/115) had no genetic alterations identified with the 24-gene panel, we next sought to determine the additional utility of applying a broader NGS panel in this cohort. We, therefore, re-analysed the FFPE tumour samples from 32 ACC patients in whom no variants had been detected on the focused NGS panels using a commercially available (>325-gene) NGS panel. Additional genomic findings were detected in 27/32 patients (84%). Figure 2 summarises the genetic alterations identified through this approach, and the full list of variations can be found in Supplementary Table S2. ESCAT scores were allocated to genes where a variant was detected; references can be found on Supplementary Table S2.Additional utility was gained in 40% (13/32) of patients with regard to being able to match them to a biomarker-stratified clinical trial.MYB-NFIB fusions and TERT promoter mutations were the most common alterations, with 22% (7/32) and 22% (7/32), respectively. There are no current compounds being developed to target TERT promoter mutations. MYB overexpression as a result of fusion transcripts is the hallmark of ACC, ATR has been shown to be downstream of MYB and in preclinical studies of ACC models, treatment with ATR inhibitors has led to apoptosis and growth inhibition [48].Genes coding for chromatin modifiers (EP300, ARID1A, KDM6A, BCOR, CREBBP, SETD2, SMARCB1) were altered in 44% (14/32) of patients, with 13% (4/32) having more than one gene altered. EP300/CREBBP inhibitors are being trialled in EP300-, ARID1A- and CREBBP-deficient tumours [49].NOTCH1 was altered in 13% (4/32), and in three cases these were activating mutations. These can be targeted by gamma secretase inhibitors, which are in phase 2 trials in ACC. AL101 achieved a response rate of 9% with a disease control rate of 70%; this currently fails to meet the ESMO Magnitude of Clinical Benefit Scale criteria and as such cannot be placed in tier 1b. As there are no data in ACC outside of NOTCH mutant patients, we cannot comment on there being a higher response rate in this setting (2b). As such, this match was given a rank of 4a [50,60].PIK3R1 was mutated in 13% (4/32) and has been shown to cause hyperactivation of the PI3K pathway [45], indicating a potential role for AKT1 inhibition. In total, 9% (3/32) had alterations in genes involved in the DNA damage repair pathway that are thought to sensitize one to PARP inhibitors [51].Previously undetected alterations in MET, ERBB2 and PTEN were detected, even though these genes were included in the primary analysis with the 24-gene panel. The alteration in MET was an amplification, which would not be detected by the 24-gene panel. The ERBB2 mutation was detected in the 24-gene panel but was not identified as being significant; this reflects the fact that there may be disagreement in approaches to variant calling. Finally, the PTEN alteration used in the secondary analysis used a tumour sample from a metastatic site, whereas the initial analysis in which this was not detected used a different sample from a primary site. This observation reflects the potential for discordance between mutations identified from primary and metastatic sites.This study sought to evaluate the clinical utility of tumour profiling by DNA-based NGS when applied to a cohort of patients with recurrent or metastatic salivary gland cancer being evaluated for clinical trials. We found, using a focused (24-gene) NGS panel, that genomic alterations were identified in 27% of patients with potential biomarker-matched clinical trial therapies available for 14% (25/175). The utility of this approach in identifying matched trial therapies was lower for ACC (7%) compared with other histological subtypes (30%), with the greatest utility seen for salivary duct carcinoma (69%). For ACC, broader (>325-gene) NGS panels provided additional utility in identifying matched clinical trial therapies, identifying matched trial therapies in 40% of patients in whom no alterations were identified using the focused panel. For this, we used a predesigned commercially available platform that is clinically and analytically validated for all solid tumours and has FDA approval as a companion diagnostic.The 24-gene panel utilized was developed for pan-tumour use, in line with the United Kingdom National Genomic Test Directory for somatic mutations in adult solid tumours. This study found that 62% (15/24) of genes in our focused NGS panel are currently being investigated for their potential value as predictive biomarkers in ongoing trials in the UK. A total of 50% (12/24) of these genes are recognised by the FDA as predictive biomarkers or standard-of-care, and as positive predictive biomarkers of response in other tumour types other than salivary gland cancer. In this study, we used the ESMO Scale for Clinical Actionability of Molecular Targets (ESCAT) to determine the level of evidence supporting the match between genomic alteration and drug trial therapy. ESCAT breaks down evidence based on trial types, with prospective trials being evidence level 1, and this is further subdivided based on whether it was randomised (1a), nonrandomised (1b) or a basket trial (1c). Furthermore, ESCAT takes into account whether there was a benefit to overall survival (1a) or overall response rate (2b) [22]. Similar scoring systems such as OncoKB can be more ambiguous on these counts, placing a heavier focus on drug and biomarker approval for use, with nonapproved drugs being classified as “showing clinical benefit”. OncoKB provides a web-based resource which allocates biomarker evidence levels automatically; however, this is of more utility in common cancers, and only provides an outline as not all genes are currently included [21].Mutations in TP53 and PIK3CA were the most frequently identified findings in this cohort. Pathogenic TP53 mutations were the most prevalent alteration identified in our focused panel, being present in 19% (34/175) of our cohort. We have previously shown TP53 mutation to be a negative prognostic factor in ACC [61]. There have, however, been no SGC specific studies investigating the predictive significance of TP53 mutations. Preclinical studies in cell lines other than SGC showed increased response rates to Wee1 inhibitors [62], but this has yet to translate into the clinic. An NCI-MPACT study, which was open to all solid tumour types, including SGC, investigated the Wee1 inhibitor Adavosertib in combination with carboplatin in a TP53-mutant cohort, but reported no significant responses [63]. Studies on ACC models have indicated that ATR inhibitors may have some benefit in ACC as ATR is downstream of MYB [48], paired with data from preclinical studies in other tumour types that have demonstrated that treatment with ATR inhibitors in TP53-mutant tumours results in increased radiotherapy and chemotherapy sensitisation [42], providing a rationale for the enrolling patients whose tumours harbour TP53 mutations in trials such as NCT03669601. However, there are more recent counterbalancing data that indicate that radiosensitisation is independent of TP53 [64].PIK3CA was the second most prevalent finding, with 11/175 patients having mutations conferring increased activity. While there have been no clinical trials investigating PI3K specifically in SGC, six ACC patients were treated with PI3K inhibitors within basket trials: 5/6 patients had stable disease at 2 months, while one patient had a partial response by RECIST 1.1 criteria. It is, however, difficult to draw many conclusions due to the small sample size and very limited follow up reported [38]. There have been no SGC-specific preclinical studies demonstrating the predictive significance of PIK3CA mutation status. PIK3CA is an established predictive biomarker for treatment with Alpelisib plus fulvestrant in hormone receptor-positive breast cancer. Alpelisib as a monotherapy in PIK3CA mutated breast cancer did not show significant clinical benefit [65]. Trials of AKT1 inhibitors are also recruiting patients with activated PIK3CA (NCT NCT01226316, NCT02338622); these are also recruiting patients with PTEN loss, and with AKT1 mutations.Although biomarker-led drug therapies are providing treatment options for patients with salivary gland cancers, this further subdivision of histopathological entities on the basis of molecular profiling exacerbates the challenges related to the rare nature of salivary cancers. Further subdividing an already small population for trial recruitment may pose a significant problem in obtaining adequate patient numbers to determine true benefits.Some of the most clinically meaningful changes associated with biomarker-directed therapy development have been reported in HER2-overexpressing/ERBB2-amplified salivary duct carcinoma with HER2-targeted therapies [6,10,11]. Now, there is clear evidence of efficacy in the recurrent or metastatic setting, and ongoing trials are now evaluating HER2-targeted therapy as an adjuvant to surgery in the curative setting (NCT04620187 [6]). In addition, in recurrent or metastatic androgen receptor-overexpressing salivary duct carcinoma, now there is clear evidence of clinical benefit from androgen deprivation therapy [7] or combined androgen blockade [8]; there is also an emerging rationale for a clinical trial of ADT/CAB in the adjuvant setting for AR-overexpressing salivary duct carcinoma at high risk of disease recurrence.Within ACC, NOTCH-activating mutations are a promising target; however, gamma secretase inhibitors such as AL101 have yet to show significant clinical benefit, being characterised by the EMSO magnitude of clinical benefit scale. However, the medium- and long-term outcomes of the phase 2 study (NCT03691207) are yet to be published. AL101 had an OR of 14% and a disease control rate (DCR) of 68% [50,60]. In NOTCH-activated ACC, which carries a significantly worse prognosis than NOTCH wildtype ACC [66,67] this DCR may translate into a meaningful clinical benefit.In addition to genomic alterations that have current clinical utility, we have also described alterations which are currently classed as level X, meaning there is a lack of evidence of actionability. TERT promoter mutations were classified as X, and were found in 22% of ACC patients sequenced with the extended panel. While it provides no current benefit for treatment selection, previous studies have shown TERT to be a positive prognostic factor [23]. BCOR was similarly classified as X; however, BCOR is frequently mutated in haematological malignancies [68] and, as such, is likely to be investigated further and may become targetable in the future. As the median survival for recurrent or metastatic ACC is in excess of 5 years, it is not inconceivable that BCOR-targeting therapies could be in the clinic in that timeframe.In this study, tumour tissue was analysed from either metastatic deposits or sites of local recurrence collected as part of their routine diagnostic or treatment. However, analysis from a single biopsy from a primary tumour or of a single metastatic site does not reliably cover the expected intra-patient tumour heterogeneity. For future studies, one approach to increase the likelihood of capturing intra-patient heterogeneity would be to include multiple biopsies or multiple time-points from individual patients. The sequencing of circulating tumour DNA (ctDNA) has been shown to be viable in the clinical setting with good concordance in variant calling with matched tumour samples. For example, the UK TARGET study identified actionable mutations in 41% (41/100) of patients through this method in a mixed cancer cohort, 25% (11/41) of which were treated in matched clinical trials (Rothwell et al.). Future prospective studies are planned in SGC which utilise sequencing of (ctDNA) to provide a minimally invasive and “current” picture of variants [69].Furthermore, tissue samples for analysis were requested to meet the minimum tumour content of 20% for analyses, for which our panel was validated to detect a variant allele frequency of 4%. However, the exact tumour content of samples was not routinely recorded for all cases which we recognize as a limitation of this study. Although the risk of omitting internal validation of tumour content is the under-calling of mutations through sequencing of an increased percentage of normal tissue DNA, our detected frequencies are however in line with publicly available datasets [19,20].While we focused on NGS though targeted gene panels within our study, whole genome (WGS) and whole exome sequencing (WES) could also be utilized in this setting. There is potential additional utility using these approaches to identify broader mutation signatures. WGS and WES also provide greater coverage capturing mutations in areas that may become clinically significant as our understanding of the underlying tumour biology increases. However, the additional data generated by WGS/WES approaches due to this broader coverage and additional sequencing steps, through sequencing matched normal DNA, require additional bioinformatics analyses. This plus the additional analytical reagents required make WGS and WES more expensive than targeted NGS panels. This contrasts with targeted NGS panels covering specific clinically important coding and noncoding regions. As such, required read depths can be achieved with minimal reagents and can be interpreted with simpler bioinformatics pipelines; however, this is at the cost of a limitation of utility for broader discovery research. The impact of mutations in genes involved in epigenetic regulation such as BCOR, can be difficult to predict due to the myriad of pathways that are regulated through epigenetics. Microarrays can be used for detecting variations in gene expression levels. This could be explored in future analyses to assess the impact of detected somatic mutations on the transcriptome. While BCOR is not currently targetable, mutations in this gene may result in overexpression a targetable protein.A further barrier to the clinical application of the genomic findings is the limited availability of clinical trials recruiting in the UK. Relatively few sites are actively recruiting to each individual trial providing a practical or geographical barrier to clinical trial recruitment. For example, there were no available clinical trials accepting salivary gland cancer patients for the approved anti-EGFR therapies, such as afatinib and erlotinib or osimertinib within the UK. The UK is lacking a large tumour-agnostic trial such as NCI-MATCH(NCT02465060) or CAPTUR (NCT03297606). However, this situation may improve as the TAPISTRY trial (NCT04589845) is set to open for recruitment in the UK, enabling access to a selection of matched trial therapies.The success of NTRK-targeted therapies in patients with secretory carcinoma harbouring NTRK gene fusions represents a significant step forward for genetic based approaches in SGC. These patients were not represented in our cohort, possibly due to the lower frequency of recurrence in this subtype in comparison to other salivary gland cancer types [70].Tumour profiling with targeted DNA panels provides valuable information to help provide a rationale for enrolling patients in early phase trials. This approach showed the most utility in salivary duct carcinoma. Expanded sequencing with >325-gene panels provided added utility in ACC and identified the deregulation of transcriptional regulation with EP300, KDM6A and ARID1A mutations, in addition to the classically reported MYB translocations.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051133/s1, Table S1: panel coverage, Table S2: ESCAT ref, Table S3: Variants.Conceptualization, R.M. and S.R. methodology, R.M. and S.R.; formal analysis, S.R. and L.W.; resources, R.M.; data curation, S.R., L.F., H.A., B.H., L.W., G.J.B., K.J.H., G.B. and R.M.; writing—original draft preparation, S.R.; writing—review and editing, S.R. and R.M.; visualization, S.R.; supervision, R.M.; funding acquisition, R.M. All authors have read and agreed to the published version of the manuscript.This research was funded by Syncona Foundation, The Harriet Bowman Foundation, The Infrastructure Industry Foundation and The Christie Charity.The study was granted research ethics approval under the MCRC Biobank Research Tissue Bank Ethics (NHS NW Research Ethics Committee 18/NW/0092) and was performed in accordance with the Declaration of Helsinki. The role of the MCRC Biobank is to distribute samples and, therefore, cannot endorse studies performed or the interpretation of results.All subjects provided informed consent to the collection of demographic, clinical, and genomic data included in this study.The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the requirement to uphold the data sharing with relevant approved researchers as stipulated in the ethical approval.The authors are grateful for the assistance of Jenni Hill and Lesley Drain for their support in co-ordinating the NGS analyses for this study.The authors declare no conflict of interest.Results of DNA-based next-generation sequencing from patients with salivary gland cancer (n = 175) using 24-gene NGS panel. (A) Adenoid Cystic Carcinoma (ACC). (B) Patients with other (non-ACC) histopathological subtypes. Other includes neuroendocrine carcinoma (n = 1), secretory carcinoma (n = 1) and NUT carcinoma (n = 1). Individual patient results are represented by a column of vertical bars. Only genes in which alterations were detected are shown. Detection of genomic alterations are indicated by coloured bars, and the absence of alterations is indicated by grey bars.Results of 325-gene NGS panel in patients where no mutation was found in the focused NGS panel. Studies providing the rationale behind ESCAT classification can be found in Supplementary Table S2 [37,39,48,49,50,51,52,53,54,55,56,57,58,59]. Selected mutations are included in the figure, including all mutations with biomarker-stratified trials; the full list is in Supplementary Table S3.Clinical characteristics of patients with salivary gland cancer undergoing genomic profiling. Abbreviations: NGS. Next-generation sequencing; ACC (Adenoid Cystic Carcinoma); Acinic (Acinic Cell Carcinoma); Adeno (Adenocarcinoma); ExPleo (Carcinoma ex Pleomorphic Adenoma); SDC (Salivary duct carcinoma); MyoE (Myo-epithelial Carcinoma); MEC (Mucoepidermoid carcinoma); NEC (Neuro-endocrine carcinoma); Secretory (Secretory carcinoma); NUT (NUT carcinoma).Components of focused NGS panel and matched drug therapies. Biomarkers were classified based on ESCAT. The ESCAT score is attributed to specific gene variants as outlined in the original papers referenced. Only evidence for point mutations and small insertions and deletions are included. Other genetic alterations such as copy number variations are not included as they were not analysed in our panel.Full list of significant alterations as classified by CanVIG [16,17]. Grouped by gene. ACC—adenoid cystic carcinoma; Acinic—acinic cell carcinoma, ADENO—adenocarcinoma not otherwise specified; ExPleo Carcinoma—expleomorphic adenoma; MEC—mucoepidermoid carcinoma; SDC—salivary ductal carcinoma.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01134.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Obesity is a rising health epidemic in breast cancer survivors and associated with multiple negative health sequalae and increased mortality. Delay Discounting (DD) is a behavioral economic measure of an individual’s valuation of future outcomes. While higher DD correlates with obesity in the general adult population, valuation of the future may impact cancer survivors differently due to their unique experiences. We assessed cross-sectional associations between DD, BMI, and healthy lifestyle behaviors in an exploratory analysis of 89 women with hormone receptor positive non-metastatic breast cancer. We found higher DD to be associated with obesity and decreased frequency of vegetable consumption. Future studies should investigate DD as a therapeutic target for novel behavioral interventions in breast cancer survivors affected by obesity. This may improve valuation of the future, increase healthy lifestyle behaviors, and facilitate weight loss to promote overall health and longevity in this population.Obesity in breast cancer (BC) survivors is associated with increased mortality. Delay discounting (DD) is a behavioral economic measure of how individuals value future outcomes. Higher DD correlates with obesity in the general population. Valuation of the future may be associated with obesity differently in cancer survivors. This study evaluated the relationship between DD and obesity in BC survivors. We report an exploratory analysis assessing cross-sectional associations between DD, BMI, and lifestyle behaviors (vegetable and fruit consumption, exercise) related to obesity in 89 women with hormone receptor positive non-metastatic BC. Multivariate linear regression analysis examined demographic and lifestyle behavior variables associated with both BMI and DD. Greater willingness to wait for larger, delayed rewards (lower DD) was significantly associated with lower BMI (standardized beta = −0.32; p < 0.01), independent of age, race, income, time since diagnosis, and menopausal status. There was no significant association between DD and fruit consumption or exercise frequency. Vegetable consumption was significantly associated with lower DD (standardized beta = 0.24; p < 0.05). Higher DD is associated with obesity and decreased frequency of vegetable consumption in BC survivors. Future studies should investigate DD as a therapeutic target for behavioral interventions to facilitate weight loss and promote longevity in this population.World-wide, breast cancer (BC) is the most commonly diagnosed cancer in women, with an estimated 2.3 million new cases in 2020 [1]. Outcomes are significantly improving in early stage BC, with an approximate five-year survival rate of 90% [2], leading to a growing population of greater than 3.8 million women in the United States who are BC survivors [3]. In addition to the risk of cancer recurrence and mortality, BC survivors also face other unique challenges associated with treatment such as obesity, depression, and a rising incidence of cardiovascular disease which can negatively impact long-term health [4]. Optimization of modifiable cancer risk factors through encouragement of healthy lifestyle behaviors is at the cornerstone of consummate cancer survivorship care.Obesity is a rising health epidemic in the United States [5] and excess adiposity is a known risk factor for BC development. Additionally, there is an increased risk of BC recurrence and mortality associated with excess body weight in BC survivors [6,7,8,9]. For example, in a meta-analysis of 82 studies including 213,075 breast survivors, a 41% increase in mortality was demonstrated in women who were obese at the time of cancer diagnoses [8]. Obesity in BC survivors is also correlated with other adverse health sequelae, including an increased risk of second primary malignancies [10], and negative impact on multiple quality of life factors, such as chronic fatigue, sexual dysfunction, body image, lymphedema, and neuropathy [11]. Additionally, there is an increased risk of chronic medical conditions known to be associated with increased adiposity such as cardiovascular disease and insulin resistance with resulting diabetes [11,12]; the presence of these comorbidities is a substantial cause of shortened survival in this population. Weight loss is, therefore, a highly impactful therapeutic target in BC survivorship. However, weight loss interventions in BC survivors have been associated with mixed outcomes in success and suboptimal adherence [13,14,15]. Novel approaches to improve long-term patient engagement and adherence of healthy lifestyle choices through identification of basic decision-making processes and intervention with behavioral modification may promote durable weight loss and its resulting long-term health benefits.Delay discounting (DD) is a behavioral economic measure that describes the degree to which individuals value future outcomes [16]. When given a choice between a larger reward in the future or a smaller more immediate reward in the present, many individuals will prefer the latter, particularly as the delay to the more beneficial outcome increases [17]. Individuals with higher, or more rapid, DD rates therefore have an increased bias for immediate gratification and a greater devaluation of future outcomes [16,18]. For example, a person with a high rate of DD may be more likely to choose to eat a delicious, but unhealthy meal now, rather than consider an option which could positively impact future fitness and health. High DD rates are implicated across many maladaptive health behaviors [19,20,21,22] in the behavioral science literature, including those related to obesity, such as sedentary behavior [23], poor glycemic control [24,25], and poor dietary choices (consumption of obesogenic foods and increased caloric intake) [26,27]. Predictably, higher DD rates are correlated with obesity [27,28]. Thus, consistent with the experimental medicine approach to behavior change research [29], DD may serve as a therapeutic target in obesity; that is, interventions that reduce DD may, in turn, facilitate weight loss. Such methods to engage DD as a target for weight loss are being implemented as an emerging treatment approach [30,31,32,33], but there is a paucity of data regarding the association between DD, obesity, and behavioral choices among those who have experienced a cancer diagnosis.Valuation of the future in cancer survivors may be altered by a number of unique factors, such as an adjusted mortality perception, financial toxicity from cancer diagnoses and treatment, and cancer-related psychosocial consequences. For instance, cancer survivors may experience psychologic effects including posttraumatic stress and a fear of cancer recurrence [34,35]. The latter may be prevalent in nearly all cancer survivors and can persist despite completion of therapy and considerable time since diagnoses [35,36]. Moreover, a traumatic experience and the resulting negative impact on overall well-being may produce a sense of foreshortened future and perception of early mortality [37]. An individual’s discount rate may also be impacted by distinct cancer-specific factors, such as diagnosis, stage, treatment history, and time since diagnosis. For example, in a study of cancer survivors of multiple cancer types, time since cancer diagnoses was negatively associated with DD rate (p = 0.01), suggesting that DD is elevated soon after diagnosis and decreases with time [38]. The multitude of these complex factors may all impact valuation of the future and potentially decrease an individual’s commitment to long-term goals and rewards related to health.Focused investigation of DD in cancer survivors is critical to better elucidate its role as a potential therapeutic target for weight loss in this population. In a study of 1001 survivors across several types of cancers, including breast, lung, sarcoma, genitourinary, and others, participants were recruited to assess associations between DD and multiple lifestyle behaviors [38]. Lower DD rates were associated with several healthy lifestyle behaviors (lower alcohol consumption, no cigarette or other tobacco use, no tanning bed use, attending annual primary care visits), highlighting the applicability of this behavioral principle to cancer survivorship. Notably, no significant interaction was found between DD rate and Body Mass Index (BMI), or between DD rate and physical activity or vegetable consumption. An important limitation of this study in specific regard to the association between DD and BC was the heterogeneous cohort of survivors, as different cancers are known to be associated with different prognoses and unique survivorship goals. A minority of patients were BC survivors (n = 175, 17.5%) and approximately half of the sample was of male gender. Notably, there was no investigation of whether cancer type could be a predictor of the interaction between DD and health behaviors. Thus, focused assessment of DD and lifestyle behaviors in BC survivors is warranted to better ascertain the relationship to adiposity, nutrition, physical activity, and specific cancer characteristics (e.g., stage, time since diagnoses).A pilot study from our investigational team was the first to study DD specifically in a BC survivor cohort and demonstrated DD as a predictor of adjuvant endocrine therapy non-adherence in 89 participants. This study found that treatment adherence was higher in participants showing greater willingness to wait for larger, delayed rewards (i.e., lower DD) (standardized beta = 0.328, p = 0.005) [39]. These findings suggest that DD does impact lifestyle behaviors in BC survivors similar to other populations and strategies to reduce DD may be efficacious to improve long-term survival. Herein, we describe additional exploratory analyses in this cohort to investigate the cross-sectional association between DD, BMI, and healthy lifestyle behaviors associated with energy balance. We hypothesized that increased BMI and associated negative lifestyle behaviors in BC survivors are associated with higher DD, similar to that observed in individuals without cancer [23,26,27,28]. Such findings could prompt further, longitudinal studies of this association, and ultimately, the development of novel behavioral strategies for weight loss in this population to address an unmet clinical need in BC survivorship.Patients were recruited from a large, community-based healthcare system in the Roanoke, VA area. Key eligibility criteria included: (1) women with stage 1–3 non-recurrent hormone receptor positive invasive BC treated with curative intent in the last five years, (2) currently prescribed or recommended adjuvant oral endocrine therapy (tamoxifen, anastrozole, letrozole, or exemestane) with or without ovarian suppression, and (3) 18–80 years old. Patients who were prescribed endocrine therapy for metastatic disease and those suffering from physical (e.g., non-ambulatory) or cognitive (e.g., dementia) impairments that may interfere with medication self-administration were excluded. All participants completed informed consent prior to study enrollment. The Carilion Clinic Institutional Review Board approved all study procedures.The primary aim of this study was to evaluate cross-sectional associations between DD and endocrine therapy non-adherence. We previously reported these findings which demonstrate that greater rates of DD were significantly associated with poorer adherence to endocrine therapy [39]. To assess the relationship between DD and obesity, we undertook several exploratory analyses, which we describe herein: to evaluate the cross-sectional correlation between (i) DD and BMI, (ii) BMI and positive lifestyle behaviors, and (iii) DD and positive lifestyle behaviors.Weight and height were collected by self-report. BMI was calculated based on the formula: BMI = weight (kg)/height (m)2. During a single 90–120 min session, participants completed a questionnaire to collect information encompassing demographics, cancer diagnoses and treatment, and healthy lifestyle behaviors. The latter included information on consumption of fruit and vegetables (number of servings per day; six categories: ranging from less than one serving a day up to five or more servings a day) as well as exercise frequency (moderate intensity physical activity; six categories: never, less often than once a year, a few times a year, a few times a month, a few times a week, daily or almost daily). Next, participants completed the five-trial, adjusting-delay task [39] to assess DD of both $100 and $1000 (order randomized). This method has been validated and is accepted in the behavioral science literature as a reliable means by which to measure DD [27,28,33]. Two amounts were examined because DD varies with the amount of reward [18]. Thus, analyses of multiple amounts provide more generalized estimates of choice. In this task, participants made repeated, hypothetical choices between a larger amount ($100 or $1000, depending on task iteration) available after a delay and half of this amount ($50 or $500) available immediately. Across trials, the delay to the larger amount is titrated based on previous choices until reaching an indifference delay (possible range: 1 h–25 years, in approximately logarithmic intervals), at which point the subjective value of both options is approximately equal. The indifference delay in this task serves as a measure of Effective Delay 50 (ED50), or the delay required for the larger reward to lose 50% of its value [39,40]. The dependent measure of DD was ED50 (in days). Longer ED50 values reflect less discounting (i.e., greater willingness to wait). ED50 values were non-normally distributed (positive skew) and were natural log transformed prior to analysis.Demographic variables are summarized using means (standard deviations), medians (interquartile ranges), and frequencies (percentages), where appropriate. Univariate linear regression analysis was performed to identify explanatory variables, including demographics and healthy lifestyle behaviors (exercise frequency, fruit and vegetable consumption), associated with BMI. To identify an optimal subset of factors associated with BMI, we used multivariate regression and performed an exhaustive model selection search and identified the final model as that with the lowest Bayesian Information Criterion (BIC). Similarly, multivariate regression with model selection were performed to predict discounting (ED50 $100 and ED50 $1000). Results are reported using standardized betas and p-values. A p-value < 0.05 was considered significant in this study.A total of 89 participants completed this study, of which details have been previously published [39]. Table 1 describes the baseline demographic, lifestyle, and clinical characteristics of the cohort. The mean age at diagnosis was 58.7 years (range: 33–77) and the mean time since diagnoses was 2.5 years. The majority (n = 58, 65%) of participants were post-menopausal and of white race (n = 85, 96%). Most individuals were with stage 1 (n = 39, 44%) or stage 2 (n = 25, 28%) BC. The mean BMI was 29.8 kg/m2 (SD 6.6). A total of 29 (33%) patients in this cohort were overweight; 38 (43%) patients were obese. Adjuvant aromatase inhibitor (anastrozole, letrozole, or exemestane) was the most common endocrine therapy prescribed (n = 53, 60%), with the remainder receiving adjuvant tamoxifen (n = 36, 40%).For DD, the mean natural log ED50 was 5.38 (1.89) and 5.69 (2.14) for ED50 $100 and $1000, respectively. For context, these values correspond to raw values of 217.02 and 295.89 days, respectively, indicating that the $100 and $1000 rewards lost half of their subjective value in approximately 7–10 months. High ED50 values (i.e., lower DD) were associated with lower BMI (ED50 $100, standardized beta = −0.32, p = 0.002; ED50 $1000, standardized beta = −0.28, p = 0.008) and this effect was independent of age, race, income, time since diagnosis, and menopausal status (Figure 1). The median frequency of healthy lifestyle behaviors included 3 (IQR 2–4) vegetable servings/day, 3 (IQR 2–3) fruit servings/day, and exercise a few times a week (IQR a few times a month to daily or almost daily). Exercise frequency (standardized beta = −0.36, p < 0.001), fruit consumption (standardized beta = −0.32, p = 0.002), and vegetable consumption (standardized beta = −0.24, p = 0.024) were significantly associated with lower BMI. Model selection identified the most probable model to include ED50 $100 (standardized beta = −0.25, p = 0.009), exercise frequency (standardized beta = −0.28, p = 0.006), and fruit consumption (standardized beta = −0.22, p = 0.029).Linear regression models were used to estimate ED50 from health behaviors. Exercise frequency (ED50 $100, standardized beta = 0.13, p = 0.242) and fruit consumption (ED50 $100, standardized beta = 0.15, p = 0.164) were not associated with DD rate; however, vegetable consumption was significantly associated with lower DD rate (ED50 $100, standardized beta = 0.24, p = 0.026). Upon model selection, only age (ED50 $100, standardized beta = 0.28, p = 0.004) and household income (ED50 $100, standardized beta = 0.38, p < 0.001) were associated with DD rate.In this study of women with non-metastatic hormone receptor positive BC, we found a significant cross-sectional association between elevated DD and increased BMI in an exploratory analysis. This relationship is consistent with previously reported findings in the general adult population affected by increased BMI [27,28]. Thus, the association between DD and obesity persists despite the unique challenges faced by individuals with a cancer diagnosis. However, it contrasts the findings of Sheffer and colleagues, who did not observe a significant association between these variables in a mixed cohort of cancer survivors [38]. This suggests that the effect of DD may be different amongst cancer survivor populations. Particularly, diagnoses of distinct cancer types may impact future thinking on choice differently. This may be related to considerable variations in prognoses, symptom burden, and quality of life based on specific cancer diagnoses. For example, BC patients have a higher cure rate and live longer compared with individuals with many other solid tumor malignancies of a similar stage [2]. Thus, research on behavioral mechanisms underlying decision making in cancer patients may benefit from studies that either focus on individual cancer diagnoses or examine a potential moderating role of diagnosis type.If the association between DD and obesity in BC survivors is confirmed in longitudinal studies, DD may be engaged as a therapeutic target, and both established and investigational behavioral therapeutic interventions have demonstrated success to decrease DD rate [42,43]. For example, one promising behavioral intervention is episodic future thinking (EFT), derived from the emerging science of prospection. In EFT, individuals generate and simulate personally meaningful and detailed future events to increase valuation of the future. The goal is to shift temporal orientation to the present [44]. EFT has successfully reduced DD rate in obese individuals [31,33,45,46,47], leading to effects of negative energy balance, including decreased caloric intake, diminished consumption of energy-dense foods, improved diet quality, and weight loss [30,31,32,33,45,48,49,50]. For example, Sze et al. observed that in an EFT study in obese participants, delivered by smartphone, there was a favorable decrease in energy intake and reduction in weight in adults following the four-week intervention compared with control [30]. Additionally, a brief EFT intervention implemented in participants of the present study demonstrated that DD is also amendable to reduction in BC survivors as compared to a control condition [39]. Based on these findings and the preliminary data presented in this report, implementation of behavioral interventions such as EFT to target DD for weight loss and promotion of positive lifestyle behaviors may also hold promise as a clinical tool in cancer survivors. These interventions may be adapted in future studies, with the goal to improve impactful clinical endpoints related to obesity in cancer survivors.This study is not without limitations. The results of this cross-sectional analysis provide insight into potential mechanisms underlying increased weight in BC survivors, but cannot determine causality. Furthermore, the objectives studied were exploratory and additional, prospective research specifically powered to assess for these effects in a cohort of BC survivors with obesity is warranted. Longitudinal assessment of adiposity collected serially would also provide a more comprehensive understanding of obesity and DD following a cancer diagnosis. Namely, changes in weight and body composition following a cancer diagnosis is not uncommon and can have important clinical implications. These additional data could inform the optimal design of behavioral interventions to target DD for weight loss. Our investigative team is currently evaluating the prospective effects of DD and related targeted interventions on weight loss after a breast cancer diagnosis (NCT05012176). In addition, we did not find a consistent, statistically significant difference between DD rate and healthy lifestyle behaviors assessed in this study. While increased vegetable consumption was associated with lower discounting, there was no significant relationship between discounting and the other lifestyle behaviors. This may be related to the small sample size. Additionally, the potential to introduce recall bias exists with the use of self-report questionnaires. Future studies should incorporate validated, higher-resolution measures to provide a more comprehensive understanding of the relationship between discounting, physical activity, and energy consumption. For example, portion size, calorie consumption, diet quality, and exercise intensity/time are measures which can provide greater precision to estimate net energy balance and may clarify the link between discounting and lifestyle behaviors. Therefore, further investigation to assess the role of physical activity and dietary factors as a potential mediator of the association between DD and obesity should be pursued using robust measurements (e.g., fitness tracking, food frequency questionnaires) and in a larger cohort of patients.The study population also lacked socioeconomic and racial diversity and the majority of participants were post-menopausal. Additionally, there was limited diversity in the BC subtype, as all participants presented with hormone receptor positive BC. Advanced BC stage was also underrepresented (4% with stage 3 BC). Given that research supports obesity is associated with multiple negative health sequalae in BC survivors regardless of receptor subtypes and stage [6,7,8,51], future studies should include a more diverse sample of these characteristics to ensure generalizability of the reported findings.Future studies should investigate the effect of toxicities related to anti-cancer therapies on the uptake of healthy lifestyle behaviors and influence on DD. This is highly relevant in the modern era of breast cancer management given the rapid expansion of adjuvant treatments in cancer survivors to decrease recurrence risk; these treatments are not without side effects. We previously reported the adverse effects and associated severity related to endocrine therapy in this patient cohort [39]. The presence of aromatase inhibitor-induced musculoskeletal symptoms should be a particular area of future research, as this toxicity could impact physical activity and associated metabolic endpoints. Additionally, future studies should evaluate the unique psychological profile of individuals, including psychiatric diagnoses and psychologic support received following a cancer diagnosis. These factors may influence personal motivation to adhere to healthy lifestyle choices and an individual’s valuation of future health outcomes [52].The current study suggests an association between DD and obesity in BC survivors and supports DD as a potential therapeutic target for weight loss. Additional research to elucidate the role of this behavioral mechanism in unhealthy lifestyle choices related to positive energy balance is indicated. Additionally, this relationship should be assessed in a more diverse BC survivorship cohort to confirm reproducibility. These findings support the potential for innovative behavioral therapies that target DD to promote weight loss and the associated long-term health benefits in BC survivors affected by increased weight. A randomized clinical trial by our research team implementing remotely delivered EFT intervention by smartphone is currently ongoing with these objectives (NCT05012176). If this strategy is feasible and efficacious to promote weight loss, improve diet quality, and reduce systemic inflammatory markers, additional investigation will be pursued across multiple cancer survivorship cohorts, as weight reduction in obesity is a pervasive target in optimal cancer survivorship care.Conceptualization, J.E.V. and J.S.; Data curation, A.T. and J.S.; Formal analysis, A.T.; Funding acquisition, J.E.V. and J.S.; Investigation, J.S., J.E.V., A.T., S.S., M.L. and J.S.S.; Methodology, J.S., J.E.V. and J.S.S.; Project administration, J.S.; Supervision, J.S.; Visualization, J.S., J.E.V., A.T., S.S., M.L. and J.S.S.; Writing—original draft, J.S.S. and J.E.V.; Writing—review and editing, J.S., J.E.V., A.T., S.S., M.L. and J.S.S. All authors have read and agreed to the published version of the manuscript.This work was supported by a pilot feasibility grant from the Center for Health Behaviors Research, Fralin Biomedical Research Institute and Carilion Clinic to foster interdisciplinary collaborations in the study of health behaviors.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of Virginia Tech Carilon Clinic (protocol number 2574, date of approval June 2018).Informed consent for this study was provided by all participants.The authors confirm that the data supporting the findings of this study are available within the article and/or the tables and figures. Additional derived data supporting the findings of this study are available from the corresponding author, Jennifer Vaughn MD, on request.The authors wish to thank the physicians of Blue Ridge Cancer Care, Decca Knight, and the Suzan G. Komen Virginia Blue Ridge leadership for their advice and assistance with recruitment. The authors would also like to acknowledge the Transdisciplinary Research in Energetics and Cancer (TREC) Training Workshop R25CA203650 (PI: Melinda Irwin).The authors have no conflict of interest to report.High discounting rate was independently associated with increased BMI (A): ED50 $100, p = 0.002; (B): ED50 $1000, p = 0.008).Baseline demographic, lifestyle, and clinical characteristics of study participants (n = 89).a Alcohol consumption based on Alcohol Use Disorders Identification Test (AUDIT) [41]; b Menopausal status was based on patient report; c Stage at diagnosis was obtained through medical records and was only available for patients recruited through physician referral (n = 68); d Aromatase inhibitor included anastrozole, letrozole, or exemestane; Abbreviation Legend: SD standard deviation; IQR interquartile range.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01135.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work as co-first authors.Current address: Praxiszentrum Lauématt, 5103 Wildegg, Switzerland.These authors contributed equally to this work as co-last authors.The screening of prostate cancer (PCa), based on the serum prostate specific antigen (PSA), is characterized by a high number of false positives, leading to overdiagnosis of healthy men and overtreatment of indolent PCa. This clinical problem severely affects the quality of life of patients, who would benefit from more specific risk stratification models. By performing a mass spectrometry (MS) screening on urine samples collected prior to prostate biopsy, we identified novel biomarkers and validated them by ELISA. Here, we show that an upfront urine test, based on quantitative biomarkers and patient age, has a higher performance compared to PSA (AUC = 0.6020) and is a feasible method to improve the eligibility criteria for prostate biopsy, to detect healthy men (AUC = 0.8196) and clinically significant PCa, thereby reducing the number of unnecessary prostate biopsies.PCa screening is based on the measurements of the serum prostate specific antigen (PSA) to select men with higher risks for tumors and, thus, eligible for prostate biopsy. However, PSA testing has a low specificity, leading to unnecessary biopsies in 50–75% of cases. Therefore, more specific screening opportunities are needed to reduce the number of biopsies performed on healthy men and patients with indolent tumors. Urine samples from 45 patients with elevated PSA were collected prior to prostate biopsy, a mass spectrometry (MS) screening was performed to identify novel biomarkers and the best candidates were validated by ELISA. The urine quantification of PEDF, HPX, CD99, CANX, FCER2, HRNR, and KRT13 showed superior performance compared to PSA. Additionally, the combination of two biomarkers and patient age resulted in an AUC of 0.8196 (PSA = 0.6020) and 0.7801 (PSA = 0.5690) in detecting healthy men and high-grade PCa, respectively. In this study, we identified and validated novel urine biomarkers for the screening of PCa, showing that an upfront urine test, based on quantitative biomarkers and patient age, is a feasible method to reduce the number of unnecessary prostate biopsies and detect both healthy men and clinically significant PCa.Prostate cancer (PCa) is one of the most frequently diagnosed cancers worldwide and a prominent reason for tumor-related deaths in men [1]. In past years, early detection of PCa and its clinical management became a controversial topic. On the one hand, implementation of the serum biomarker prostate specific antigen (PSA), as a standard for the screening of PCa in the early 1990s, resulted in an increased diagnosis of early-stage tumors and a reduction of PCa-specific mortality rates [2]. Additional refinements in the PCa screening procedure due to new biomarkers and technologies, such as magnetic resonance imaging (MRI), have further improved the predictive performances of PSA [3]. On the other hand, specificities of current diagnostic examinations remain low and still lead to a high number of false positives, resulting in unnecessarily performed prostate biopsies [4]. Therefore, overdiagnosis of healthy men and overtreatment of indolent PCa remains a clinical challenge with significant impact on the quality of life of patients due to possible severe side effects [5,6]. To overcome this problem, more specific risk stratification models that can complement PSA testing need to be developed, to distinguish clinically significant from indolent PCa, and to reduce the number of biopsies performed.Urine is an ideal clinical specimen for diagnostic testing. Its easy collection is completely non-invasive and it allows the processing of large volumes, compared to tissue, blood or other biological materials. This enables the detection of biomarkers at any time point during patient care and facilitates not only diagnosis, but also the monitoring of the disease. The detection of biomarkers in urine has been studied for a wide range of cancers with ultrasensitive screening methods, such as nuclear magnetic resonance (NMR) spectroscopy and mass spectrometry (MS) [7,8]. Specific metabolites were examined for their potential to screen for cancers of the urological system, but also for non-urological tumors such as lung, breast, colorectal, gastric, hepatic, pancreatic, and renal cancer [9].The prostate epithelium secretes cellular substances into the gland and prostate cancer cells can be shed into the prostatic fluids, where they exude into the urine [10,11]. Sensitive assays can then detect DNA, RNA, proteins, and exosomes of tumor origin [12,13]. MS proteomics can be a powerful tool for high-throughput screening of proteins in urine and can be used for the identification of new biomarkers [14,15]. The translation of such methods into the clinic for standard diagnostic screening is elusive because of the high cost of instruments and the need for specially trained personnel. Therefore, validation studies of biomarkers are often performed on larger patient cohorts with immunological assays such as ELISA, which is a well-established method for protein quantification.The aim of this study was to discover novel urine biomarkers for the detection of PCa and investigate their potential as an improved diagnostic test. The goal was to select, with high sensitivity, men with unspecifically elevated PSA from men who could benefit from prostate biopsy, which remains the standard of care for the diagnosis of PCa. Since low-grade PCas are generally considered indolent, the aim of the study was also to identify biomarkers for the selection of men harboring high-grade PCa. Thus, by improving the eligibility criteria for prostate biopsy, we would reduce the number of unnecessary prostate biopsies performed. Additionally, it might offer the possibility of non-invasive disease monitoring. Tests that rely on the quantification of single biomarkers are often limited in their power to predict cancer, a disease that is hallmarked by its heterogenic biology [16,17]. Therefore, we focused on the quantification of multiple biomarkers to achieve an increased accuracy in predicting PCa.We performed a MS screening on urine samples from 45 men with elevated PSA levels scheduled for prostate biopsy and identified 2.735 proteins across all samples, as well as potential biomarkers for the detection of all grades of PCa or high-grade tumors only. Top candidates were then validated by ELISA and a combinatory analysis predicted their performances as multiplexed diagnostic test for PCa screening.A total of 45 patients were enrolled in the study at the Urology Department of the University Hospital of Zürich (Zürich, Switzerland). Samples were collected as first-morning urine from men not subjected to prostatic massage, with high serum PSA levels (≥2 ng/mL) and/or abnormal digital rectal examination (DRE) results, before the performance of the prostate biopsy. Sample aliquots were then stored at −80 °C until use. Patients’ recruitment, urine sample collection, and analysis were approved by the Ethics Committee of Kanton Zürich (BASEC n° 2016-00829).Mass spectrometry (MS) analysis was performed by Biognosys AG (Schlieren, Switzerland). All solvents were HPLC-grade from Sigma Aldrich (Schaffhausen, Switzerland) and all chemicals, if not stated otherwise, were obtained from Sigma Aldrich (Schaffhausen, Switzerland).After thawing, sample digestion was performed on single filter units (Sartorius Vivacon 500, 30.000 MWCO HY) following a modified FASP protocol (described by the Max Planck Institute of Biochemistry, Martinsried, Germany). Samples were denatured with Biognosys’ Denature Buffer and reduced/alkylated using Biognosys’ Reduction/Alkylation Solution for 1 h at 37 °C. Subsequently, digestion to peptides was carried out using 1 μg trypsin (Promega) per sample, overnight at 37 °C.Peptides were desalted using C18 Ultra Micro Spin columns (The Nest Group) according to the manufacturer’s instructions and dried down using a SpeedVac system. Peptides were resuspended in 17 μL LC solvent A (1% acetonitrile, 0.1% formic acid (FA)) and spiked with the Biognosys iRT kit calibration peptides. Peptide concentrations were determined using a UV/VIS Spectrometer (SPECTROstar Nano, BMG Labtech, Ortenberg, Germany).For HPRP fractionation of peptides, digested samples were pooled. Ammonium hydroxide was added to a pH value > 10. The fractionation was performed using a Dionex UltiMate 3000 RS pump (Thermo Scientific™) on an ACQUITY UPLC CSH C18 1.7 μm, 2.1 × 150 mm column (Waters). The gradient was 1% to 40% solvent B in 30 min, solvents were A: 20 mM ammonium formate in water, B: acetonitrile. Fractions were taken every 30 s and sequentially pooled to 12 fraction pools. These were dried down and resolved in 15 μL solvent A. Prior to mass spectrometric analyses, they were spiked with Biognosys’ iRT kit calibration peptides. Peptide concentrations were determined using a UV/VIS Spectrometer (SPECTROstar Nano, BMG Labtech).For shotgun LC–MS/MS measurements, 2 μg of peptides per fraction were injected to an in-house packed C18 column (Dr. Maisch ReproSil-Pur, 1.9 μm particle size, 120 Å pore size; 75 μm inner diameter, 50 cm length, New Objective) on a Thermo Scientific Easy nLC 1200 nano-liquid chromatography system connected to a Thermo Scientific™ Q Exactive™ HF mass spectrometer equipped with a standard nano-electrospray source. LC solvents were A: 1% acetonitrile in water with 0.1% FA; B: 15% water in acetonitrile with 0.1% FA. The nonlinear LC gradient was 1–52% solvent B in 60 min followed by 52–90% B in 10 s, 90% B for 10 min, 90–1% B in 10 s and 1% B for 5 min. A modified TOP15 method from Kelstrup was used [18]. Full MS covered the m/z range of 350–1650 with a resolution of 60.000 (AGC target value was 3 × 106) and was followed by 15 data dependent MS2 scans with a resolution of 15.000 (AGC target value was 2 × 105). MS2 acquisition precursor isolation width was 1.6 m/z, while normalized collision energy was centered at 27 (10% stepped collision energy) and the default charge state was 2+.For DIA LC–MS/MS measurements, 2 μg of peptides and 1 IE of PQ500 reference peptides were injected per sample. For samples with less than 2 μg of total peptide available, the amount of reference peptides was adjusted accordingly. Peptides were injected into an in-house packed C18 column (Dr. Maisch ReproSil-Pur, 1.9 μm particle size, 120 Å pore size; 75 μm inner diameter, 50 cm length, New Objective) on a Thermo Scientific Easy nLC 1200 nano liquid chromatography system connected to a Thermo Scientific Q Exactive HF mass spectrometer equipped with a standard nano-electrospray source. LC solvents were A: 1% acetonitrile in water with 0.1% FA; B: 15% water in acetonitrile with 0.1% FA. The nonlinear LC gradient was 1–55% solvent B in 120 min followed by 55–90% B in 10 s, 90% B for 10 min, 90–1% B in 10 s, and 1% B for 5 min. A DIA method with one full range survey scan and 22 DIA windows was used.The shotgun mass spectrometric data were analyzed using Biognosys’ search engine SpectroMine™, the false discovery rate on peptide and protein level was set to 1%. A human UniProt FASTA database (Homo sapiens, accessed on 1 July 2019) was used for the search engine, allowing for two missed cleavages and variable modifications (N-term acetylation, methionine oxidation, deamidation (NQ), carbamylation (KR)). The results were used for generation of a sample-specific spectral library.HRM mass spectrometric data were analyzed using Spectronaut™ 14 software (Biognosys). The false discovery rate (FDR) on peptide and protein levels was set to 1% and data were filtered using row-based extraction. The spectral library generated in this study was used for the analysis. The HRM measurements analyzed with Spectronaut™ were normalized using global normalization.For testing of differential protein abundance, MS1 and MS2 protein intensity information was used [19]. Protein intensities for each protein were analyzed using a two sample Student’s t-test, and p-values were corrected for overall FDR using the q-value approach [20]. The following thresholds were applied for candidate ranking: q-value < 0.05 and absolute average log2 ratio > 0.8074 (fold change > 1.75). After removal of proteins that were not identified in at least 90% of the samples, a selection based on ROC analysis was performed in order to identify the final list of the best performing 25 candidates (AUC > 0.670 and >10% specificity at 100% sensitivity).Validation of mass spectrometry results was performed using commercially available ELISA kits and following the manufacturers’ protocols (Table S1). Before use, urine sample aliquots were equilibrated to room temperature. Measurements were conducted using the Epoch 2 microplate reader (BioTek, Zürich, Switzerland) and data were analyzed with the Gen5 software (version 2.09, BioTek, Zürich, Switzerland).For immunohistochemical evaluation a representative tissue block of n = 11 prostate adenocarcinoma cases, including periurethral tumor manifestations if available, was selected and stained for specific antibodies (Table S2). Staining and detection was performed using an automated staining system (Ventana). Semi-quantitative evaluation for each antibody was performed by two experienced pathologists. For each tissue block a corresponding hematoxylin–eosin (HE)-stained slide was available for morphological identification of prostate cancer. For each immunohistochemical marker the expression in the tumor and normal prostatic tissue were evaluated separately by assigning a four-tiered score (0 = negative, 1 = weak, 2 = moderate, 3 = strong). The extent of stained benign and malignant glands was estimated in 10% increments. In addition, the cellular compartment of the staining for both tumor area and normal prostatic glands was specified, whereas in the normal prostatic glands further evaluation of the distinct stained cell type (luminal and basal cells) was recorded. The predominant staining pattern was assessed when considerable heterogeneity of the staining intensity was detected.All statistical analyses (except for mass spectrometry data) were performed with the GraphPad prism software, version 9. Continuous variables were expressed as box-plots (from the 25th to the 75th percentile and median), with whiskers representing the minimum and the maximum values. Statistical significance was calculated with the unpaired non-parametric Mann–Whitney U test.For the characterization of single biomarkers, ROC curve analysis was performed applying the Wilson/Brown method, whereas for combinatorial analysis of non-correlated proteins, a multiple logistic regression was applied. The correlation matrix was assessed with the Pearson correlation method.An online tool was used to draw volcano plots (VolcaNoseR, https://huygens.science.uva.nl/VolcaNoseR/, accessed on 8 September 2021).A total of 45 consecutive men with suspected PCa were enrolled in this study and underwent a prostate biopsy after urine sample collection. Their demographic and clinical characteristics are summarized in Table 1, including age, serum PSA and prostate volume. Biopsy results are classified according to the Gleason score (GS) and evaluated for diagnostic purposes by genitourinary pathologists at the University Hospital of Zürich. PCa was detected in 46.7% (21/45) and clinically significant PCa (GS 7–9) in 37.8% of the patients. More precisely, 8.9% of the patients were diagnosed with GS 6, 17.8% with GS 7a/b, and 20.0% harbored a GS 8 or GS 9 tumor. Gleason score follow-up at repeated biopsies or upon prostatectomy showed that only one patient was upgraded.Collected urine samples were then screened by MS and potential novel biomarkers analyzed by ELISA (Figure 1A).For mass-spectrometry, a spectral peptide library was generated by shotgun LC–MS/MS of high-pH reversed-phase chromatography (HPRP) fractions from all 45 urine samples. Two samples showed a significant contamination with albumin, which led to the suppression of other peptide signals, and were therefore excluded from further analysis (data not shown). We identified a total of 38.454 precursors (peptides including different charges and modifications), corresponding to 23.059 unique peptides and 2.768 proteins across all 43 urine samples by using a false discovery rate of 1% (Figure 1B).For the identification of candidate biomarkers to detect healthy men, we compared the abundance of 2.768 proteins in samples from patients not affected by tumor and those with PCa. Significantly dysregulated proteins were identified by setting the q-value below 0.05, at an average fold change of more than 1.75, resulting in 351 biomarker candidates (Figure 1C, Table S3). Strikingly, most of the candidates (321) displayed decreased levels in the urine of PCa patients compared to healthy men. In contrast, only 30 candidate biomarker candidates were found to have increased levels in the “tumor” group.A key selection criterion for the best target molecules from the screening was the ability to discriminate healthy patients (with high specificity and accuracy), achieving a negligible number of false negatives (sensitivity > 90%). For this reason, all proteins that were not detected in more than three samples were excluded from further analysis. Additionally, proteins with low diagnostic performances, displaying a receiver operating characteristic (ROC) area under the curve (AUC) smaller than 0.670 and a specificity of less than 10% at 100% sensitivity, were removed. This ranking resulted in 43 biomarkers, with the top 25 candidates listed in Table S4. Among them, pigment epithelium-derived factor (PEDF), hemopexin (HPX), cluster of differentiation 99 (CD99), calnexin precursor (CANX), FCER2 (CD23, Fc fragment Of IgE receptor II), hornerin (HRNR), and keratin 13 (KRT13) showed remarkable diagnostic performance (Figure 2A,B; Table 2) and were selected for further validation by means of commercially available ELISA kits. Notably, all these biomarkers showed decreased levels in patients harboring prostate cancer.The illustrated box plots in Figure 2A show the intensities of the biomarkers in patients with and without PCa as quantified by MS. All biomarkers identify true negative patients that could be spared from performing an unnecessary prostate biopsy, although the p value was a borderline result in terms of statistical significance for two biomarkers. The ROC plots (Figure 2B) show the ability of the single biomarkers to detect all PCa (GS 6–9, red curves) in comparison to the current standard of care, which is serum PSA (black curves). Each of the seven biomarkers had a superior performance compared to PSA and was able to correctly classify 100% of patients with PCa, while detecting tumor free men at varying specificities (Table 2).Taken together, these data demonstrate that urine is a reliable proteomic source of biomarkers for the early detection of PCa and that the seven selected biomarker candidates are capable of sparing a relevant number of men from unnecessary prostate biopsy while avoiding misdiagnosis of patients bearing a prostate tumor.To assess potential biomarker combinations via multiple logistic regression, we first performed a Pearson correlation analysis among biomarker levels in the patient cohort (Figure 2C). In fact, the combination of variables can improve the performance of a predictive model only if the variables are not correlated to each other. In our analysis, we therefore combined biomarkers with a correlation coefficient of up to 0.3. Since the size of the cohort is limited to 43 patients, combinations of a maximum of two biomarkers were taken into consideration, in order to prevent the generation of overfitted models. All possible 14 combinations of biomarkers revealed a significantly larger AUC compared to the null hypothesis of AUC = 0.5 (Table 2). Moreover, any combination of two proteins led to a superior diagnostic performance, with increased AUC and higher specificity at 90% and 100% sensitivity compared to the single biomarkers. As an example, Figure 2D illustrates the multiple logistic regression curve of the PEDF and FCER2 combination (red line), which reached the best specificity of 72.7% at 100% sensitivity. This indicates that potentially 72.7% of healthy men could be spared from performing an unnecessary biopsy.Our data show that the combination of biomarkers markedly improves the diagnostic power of the model and leads to the superior detection of healthy patients who could be spared from a prostate biopsy.The validation of the candidate proteins selected from the MS analysis was performed by ELISA. Conversely to MS, immunoassays are standardized techniques that can be easily performed in any laboratory and allow for easy comparison among cohorts. For the MS measurements, the different urine samples were normalized according to their total peptide concentration and a defined amount of 2 µg was injected for each run. This approach cannot be applied to ELISA. Nevertheless, normalization is necessary to compensate for variations due to diet, time of collection and physiological characteristics of patients. Therefore, we have chosen non-dysregulated molecules from the mass-spectrometry analysis, i.e., cluster of differentiation 44 (CD44) and ribonuclease A family member 2 (RNASE2) and used them as controls for ELISA quantification of the single biomarkers (Figure S1; Table S4). Consistent with the corresponding MS data, Mann–Whitney U analysis of the normalized ELISA data for each analyte showed a significant difference between patients diagnosed with PCa and healthy individuals (Figure 3A). Furthermore, ROC curve analysis is concurrent with each MS dataset, demonstrating that all biomarkers have the diagnostic potential to detect healthy men at 100% sensitivity (Table 3).Detection of high grade PCa has a relevant clinical impact, as it allows differentiation between patients who would benefit from active surveillance and those who need active treatments. We therefore also tested the potential of our biomarkers to discriminate also PCa GS ≥ 7. The quantitative analysis by ELISA shows that the seven biomarkers can detect high-grade PCa with high performance (Figure 3B, Table 3).When different biomarkers are normalized by the same controls, as in this study, their combinatory power is hampered by a highly correlated dataset (data not shown), driven by the identical normalization strategy. Hence, combinatorial analysis was performed by multiple logistic regression with non-normalized ELISA data. In this study, we excluded from the nomogram any clinical and demographic information with potentially high variability among individual clinics and cohorts. Prostate volume and digital rectal examination (DRE), for example, are known to be affected by the type of instrument used or by personnel expertise. We therefore included only the age of the patients as clinical variable to improve the predictive models. The Pearson correlation analysis of all variables is shown in Figure 4A. All combinations, including age, resulted in a significantly higher AUC compared to the null hypothesis and were able to detect all grades of PCa with 100% sensitivity (Table 4). As an example, the ROC curve of two of the best performing combinations, PEDF + FCER2 + age and KRT13 + FCER2 + age showed a specificity of 39.1% and 52.2% at 100% sensitivity, respectively (Figure 4B). Moreover, for the detection of high-grade tumors, the combination of uncorrelated analytes increased the overall performance of the single biomarkers. As model example, the ELISA quantification of KRT13, FCER2 + age showed a striking AUC of 0.7801 with a specificity of 48.1% at 100% sensitivity (Figure 4C).Taken together, our data demonstrate that ELISA quantification of the biomarker candidates selected by MS is feasible and confirms the high diagnostic performance of the analytes, both as single and in combination for the detection of all PCa grades and clinically significant tumors (GS ≥ 7).To investigate the possible origin of the biomarkers, we performed immunohistochemistry analysis on prostate tissues from 11 men (of the initial 45 patients) that underwent radical prostatectomy. Because it was not possible to analyze prostate tissue from healthy patients, the healthy tissue areas of the prostate were used as control for each patient who underwent prostatectomy. The stainings were performed on tissue blocks, including benign and malignant areas of the prostate to compare biomarker expression levels. In concordance with the MS and ELISA data, KRT13 staining showed a distinct expression in benign and low expression in malignant tissue areas (Figure 5A,B; Table S6). We observed basal cell staining for KRT13, PEDF, and HPX in benign regions of the gland, a cell type that is absent in acinar-type adenocarcinomas (Figure 5A–F). Immunohistochemical analysis of CD99, HRNR, and CANX confirmed the expression of these markers in the prostate but, due to high heterogeneity, with high- and low-expression areas in both healthy and tumor tissues, it was not possible to compare the two conditions (Figure S2). No expression of the B-cell specific antigen FCER2 was detected in the prostate (Figure S2).Despite continuous improvements in the reduction of overdiagnosis and overtreatment of men suspected of having PCa, the number of healthy men that are subject to invasive procedures remains high [6,21]. This trend is concordant with our cohort. For this study, patients were selected for prostate biopsy only due to abnormal DRE results and/or elevated PSA levels. Approximately half (53.3%) of patients resulted having no tumor and should have been spared from performing the biopsy (Table 1).Thus, the aim of this study was to identify novel urine biomarkers to improve the eligibility criteria for prostate biopsy and to more specifically discriminate PCa at an early stage, reducing the number of unnecessary biopsies. Here, we demonstrated the feasibility of diagnostic tests for the screening of PCa relying on urine biomarkers that can be routinely quantified by standardized laboratory methods such as ELISAs.Urine samples were collected from patients before performing the biopsy and subjected to proteomic screening by mass-spectrometry (MS) to select biomarker candidates that are dysregulated when a prostate tumor is present. Although MS results showed promising results, the application of mass-spectrometry for urine analysis as routine diagnostic test is not feasible, due to the lack of a standard method to compare different batches of samples. A more practical approach is the implementation of quantitative immune-assays such as ELISA, which represents the gold standard for biomarker assessment and validation [22]. Consequently, among the 25 most performant candidates, seven proteins (PEDF, HPX, CD99, FCER2 (CD23), CANX, HRNR, and KRT13) were subsequently quantified in the same urine samples by quantitative ELISA. Additionally, their performance for the diagnosis of PCa and prediction of high-grade tumors was assessed. Although the translation of targeted MS assays into the clinical diagnostic setting appears to be difficult due to high costs and specific expertise requirements [23], the validation by ELISA demonstrates the feasibility of a clinical implementation through standard techniques. MS results of the 25 top ranked biomarkers in this study showed a significant decrease in signal intensity when a prostate tumor is present and can identify PCa patients with better performance compared to the standard PSA test (Table S4).PEDF showed the best performance as a single biomarker, with AUC of 0.8023 and specificity of 36.4% at 100% sensitivity (Figure 2A,B). On the other hand, as an example of the many possible options (Figure 2D), the best performing combination of PEDF and FCER2 markedly increase the AUC in predicting PCa compared to each individual marker and also to PSA. Specifically, with this combination 72.7% of unnecessary biopsies could be avoided, without missing any patient with PCa (100% sensitivity).The proteomic content of urine is affected by many factors, such as individual life-style, diet and time of sampling. For this reason, absolute biomarker data need to be normalized with a different strategy compared to MS, in which normalization is based on the overall cohort protein content. Figure 3A shows normalized ELISA results of the biomarkers panel, where each single molecule shows a strong diagnostic performance, in concurrence with the MS data. By combining KRT13 and FCER2 with age, we reached an AUC of 0.8196 and a specificity 52.2% at 100% sensitivity (Figure 4B). Besides the early detection of PCa, risk stratification of patients to better select clinically significant tumors is important to support optimal treatment options. For this reason, we have assessed the ability of the seven biomarkers to also detect tumors with GS ≥ 7 as well. Figure 3B shows that all candidates can predict the presence of high-grade PCa more precisely than serum PSA. The combination of KRT13 and FCER2 with age for the detection of high-grade PCa reached an AUC of 0.7801 and a specificity of 48.1% at 100% sensitivity (Figure 4C), thus potentially reducing the number of unnecessary biopsies almost by half, without missing any patient with clinically relevant PCa. Depending on the clinic, region and patients’ characteristics (e.g., age and expectation of life), men with low grade PCa (GS 6) will either be monitored or treated by local therapy options. In both cases, the novel biomarker panel can be applied to reduce unnecessary biopsies and monitor patients continuously and non-invasively. Therefore, by combining different biomarkers, we observed a relevant reduction of unnecessary biopsies, either performed on healthy individuals or on patients affected by clinically indolent tumors.A relevant portion of the proteins identified in our study has already been described in other mass-spectrometry analyses of urine and to a lesser extent, in urinary extracellular vesicles, plasma or prostate tissue of patients. The seven biomarkers validated in our study were chosen exclusively based on their ability to predict PCa prior to biopsy and not considering their biological function. Nevertheless, some of them have been reported to be related to cancer. Although signal reduction in case of tumor progression as described for the seven biomarkers might be surprising, both literature and tissue analysis performed in this study support these findings. Hornerin (HRNR), a member of the fused-type S100 protein family, was shown to be expressed and to play a role in different tumor types [24,25,26]. Other members of the same protein family were examined in prostate tissue of PCa patients, demonstrating that the loss of S100A2 and increased expression of S100A4 are hallmarks of PCa progression [27]. Similarly, the prostate tissue analysis of the pigment epithelium-derived factor (PEDF), a natural angiogenesis inhibitor in prostate and pancreas [28,29], showed minimal expression in high grade PCa (GS 7–10), in contrast to healthy prostate tissue, where the staining shows high intensity [28]. The downregulation of CD99 was already shown to be essential for tumorigenesis. This has been described for several tumors [30,31,32], including prostate cancer [33]. In fact, the overexpression of CD99 in prostate cancer cells inhibited their migration and metastatic potential in both in vitro and in vivo experiments [31]. Hemopexin (HPX) has been described to be downregulated in urine from PCa patients compared to tumor free men, an observation that is in concordance with our findings [34]. Moreover, a bioinformatics analysis of multiple urinary and tissue proteomes revealed HPX downregulation in high-grade PCa compared to healthy tissue [35]. In contrast to our results, elevated levels in cancer have been reported for the remaining molecules. Increased levels of the Fc fragment of IgE receptor II (FCER2) have been implicated in different hematological malignancies and sarcomas [36,37,38,39,40,41]. In addition, FCER2 is expressed in subsets of B cells and in particular depicts follicular dendritic cell networks [42], whereas expression changes in urine could reflect an altered immune microenvironment in prostate adenocarcinoma patients. Keratin 13 (KRT13) belongs to the type I keratin family and its reduced expression has been associated with oral squamous cell carcinoma lesions [43,44,45] and bladder cancer [46]. In contrast to our results, a study in 2016 revealed a correlation between KRT13 tissue expression and prostate cancer metastasis [47]. However, as we could show expression of KRT13 in the basal cells of benign glands, and since the loss of basal cells is one hallmark of prostate adenocarcinoma [48], lower expression levels in urine could also be explained by increased tumoral occupation of the gland. The endoplasmic reticulum chaperone calnexin (CANX) is associated with newly synthesized glycoproteins and involved in correct protein folding [49]. So far, CANX has not been described in PCa but its altered expression has been associated with other cancers [50,51]. To the best of our knowledge, this is the first study to suggest a putative role in PCa for the above-described biomarkers in PCa, demonstrating their dysregulation at such an early stage (prior to biopsy) and the feasibility of their quantitative assessment in urine.To investigate the possible origin of the biomarkers and their route to the urine, we performed a sequence-based analysis, predicting secretion pathways of proteins with the SecretomeP 2.0 server (http://www.cbs.dtu.dk/services/SecretomeP/, accessed on 5 October 2021). PEDF, HPX, CD99, and CANX are expressed with signal peptides and potentially traffic through the classical pathway (Golgi apparatus), whereas membrane protein FCER2 was predicted to traffic through a non-classical pathway. Conversely, KRT13 and HRNR do not appear to be secreted. This suggests that the proteins detected may be present in urine due to either the presence of cellular debris or particles deriving directly from the prostate or through blood filtration.The prostate tissue analysis performed in this study confirms that six out of seven biomarkers validated by ELISA are expressed in prostatic adenocarcinomas. Intensity analysis shows that KRT13 levels are lower in tumor tissue compared to healthy prostate, in agreement with the MS and ELISA data. Tissue staining further revealed that KRT13, PEDF, and HPX are predominantly expressed in basal cells of the benign tissue, whereas they are not detected in tumor areas where basal cells have been lost. Notably, these findings are in support of the decreased levels detected in urine of PCa patients, as the basal cells might be responsible for the direct shedding or secretion of these biomarkers into the acinar lumen and thus the loss of expression of the biomarkers can be reflected in their dysregulated levels detected in the cohort. The heterogeneous expression of CD99, HRNR, and CANX in both healthy and tumor tissue hampered the quantitative comparison. FCER2 was not detected in prostate tissue and might derive from immune cells, as it is known to be expressed in B lymphocytes [52], thus suggesting that a relevant involvement of the immune system in PCa could be detected in urine at an early stage.The present study has some limitations. First, it is a retrospective and single institution based study. Second, it relies on a small sample size, combining data of 43 patients for biomarker identification and validation. This became particularly evident when performing the multiple logistic regression analysis, as the cohort size determines the number of variables that can be combined to improve the model. To avoid false associations and large standard errors, a minimum number of five to ten events per predictor variable (EPV) has to be considered [53]. Since our cohort comprises 23 healthy men, we included no more than two to four predictor variables. Future studies investigating larger cohort sizes will allow the inclusion of higher numbers of variables and thereby improve their diagnostic performance. Nevertheless, for an explorative analysis of the biomarker candidates, the cohort provided a sufficient sample size and the combination of two to three variables yielded robust prediction models. Although it was currently not possible to validate the biomarkers in an independent cohort, their performance in this study was proved by use of two different and independent quantitative technologies, and the concordance of the findings underscores the importance of further validation of the targets.In conclusion, here, we demonstrated that an upfront urine test based solely on the quantification of novel biomarkers is a feasible approach to improve eligibility criteria for a prostate biopsy and to detect the presence of high-grade PCa, independent of serum PSA, digital rectal examination, and clinical variables. The clinical implementation of a simple urine test represents one possible and safe way to reduce the overdiagnosis and overtreatment of PCa. Furthermore, since it is completely non-invasive, it could potentially be used for disease monitoring and active surveillance.This study was submitted for patent application (applicant: University of Zürich; inventors: I. Banzola, N. Alijaj, B. Pavlovic, D. Eberli). The patent application was submitted to the European patent office (application number: EP 21/215742.4).The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051135/s1, Figure S1: Mass spectrometry analysis of two possible control molecules; Figure S2: Representative images of HRNR, CD99, CANX and FCER2 immunohistochemical stainings in one prostate adenocarcinoma patient (10× objective); Table S1: Commercial ELISA kits used for the validation of biomarker candidates; Table S2: Antibodies used for the immunohistochemical staining of prostate tissues; Table S3: Ranked candidate biomarkers from the MS screening for the detection of PCa; Table S4: Top 25 biomarkers and two control molecules resulted from mass spectrometry screening; Table S5: Statistical analysis of the biomarkers’ mass-spectrometry and ELISA quantification results; Table S6: Immunohistochemical staining of eleven prostate adenocarcinoma cases.N.A., B.P., J.H.R., M.D.B. and N.J.R. acquired data. B.P. prepared the urine samples for mass spectrometry and ELISA measurements. N.A. and B.P. performed the ELISA experiments. M.D.B. and N.J.R. evaluated the immunohistochemical stainings of prostate tissues, which were comprehensively reviewed by J.H.R., N.A., B.P., M.D.B., J.H.R. and N.J.R., with contributions from D.E. and I.B. The collection of samples and clinical data were performed by A.R., K.S., C.P., T.H., P.O., V.C., M.P. and M.V. (Markus Veit). Substantial contributions to the conception, design, and intellectual content of the paper was made by N.A., B.P., P.M., L.D., M.V. (Massimo Valerio), D.E. and I.B. The paper was written by N.A., B.P. and I.B.; N.A. and B.P. contributed equally to this work as co-first authors; I.B and D.E. contributed equally to this work as co-last authors. All authors have read and agreed to the published version of the manuscript.This work was supported by the “Gebert Rüf Stiftung” (GRS-039/18), the “Innosuisse-Swiss Innovation Agency” (40242.1 IP-LS), the “UZH Entrepreneur Fellowship” (University of Zurich, MEDEF 20018), and the “BRIDGE Proof of Concept programme” (Swiss National Science Foundation and Innosuisse, 40B1-0_203684/1).The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of Canton Zürich, Switzerland (BASEC n° 2016-00829).Informed consent was obtained from all subjects involved in the study.All data presented in this study are available in the manuscript and in the supplementary materials. Additional information are available for bona fide researchers who request it from the authors.The authors would like to thank the SNSF and the Innosuisse programme for their support and Alexandra Veloudios for the collection of urine samples and the assistance in providing patient data. We are obliged to all patients for their dedicated collaboration.N.J.R. discloses an advisory board function and receipt of honoraria from F. Hoffmann-La Roche AG. This study was submitted for patent application (applicant: University of Zürich; inventors: I. Banzola, N. Alijaj, B. Pavlovic, D. Eberli). The patent application was submitted to the European patent office (application number: EP 21/215742.4).Identification of candidate urine biomarkers by mass spectrometry. (A) Schematic workflow overview of urine biomarker screening via mass spectrometry and validation with ELISA; (B) 2.768 proteins, 23.059 peptides, and 38.454 precursors were quantified across all 43 urine samples. (C) Volcano plot of 2.768 proteins quantified by mass spectrometry. The 351 differently distributed protein candidates are shown in blue (decreased in tumors) and red (increased in tumors) and were defined by: q-value < 0.05 and average fold change > 1.75. The seven candidates PEDF, HPX, CD99, CANX, FCER2, HRNR, and KRT13 are indicated.Potential candidate biomarkers for the detection of healthy men. Mass-spectrometry based quantification of the biomarkers (A) PEDF, HPX, CD99, CANX, FCER2, HRNR, and KRT13 in patients with and without PCa. Results are expressed as box-plots (from the 25th to the 75th percentile and median) with whiskers representing the minimum and the maximum values (Table S5). Statistical difference was assessed by the unpaired non-parametric Mann–Whitney U test with p ≤ 0.05 defined as statistically significant (ns p > 0.05; * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001). (B) Diagnostic performances of the selected biomarkers assessed with the receiver operating characteristic (ROC). Each single biomarker (red curve) has a higher performance compared to serum PSA (black curve, AUC = 0.6020). (C) Correlation matrix assessed with the Pearson correlation method showing the correlation coefficients of the seven biomarkers with each other. A correlation between variables is defined as low for values up to ±0.3, medium for values up to ±0.5 and large for values up to ±1. (D) Combinatory analysis of non-correlating biomarkers via multiple logistic regression for the identification of tumor-free men. Coupling of PEDF and FCER2 resulted in the best performing biomarker combination, with an AUC of 0.8773 and a specificity of 72.7% at 100% sensitivity. Combined biomarkers displayed a higher performance compared to the single candidates and to serum PSA (black curve, AUC = 0.6020).Validation of candidate biomarkers with ELISA for the detection of healthy men or high-grade PCa. Commercially available ELISA kits were used and results for PEDF, HPX, CD99, CANX, FCER2, HRNR, and KRT13 are represented as box-plots, where the relative concentration of the biomarkers normalized to two control molecules (CD44 and RNASE2) is compared for men with (A) no tumor to patients with any grade of PCa and (B) men with no tumor or low grade (GS = 6) PCa to patients harboring a high-grade tumor (GS ≥ 7). Significance was assessed with a statistical Mann–Whitney test (ns p > 0.05; * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001). Results are expressed as box-plots (from the 25th to the 75th percentile and median) with whiskers representing the minimum and the maximum values (Table S5). The diagnostic potential of the single biomarkers was investigated with receiver operating characteristic (ROC) analysis. All biomarkers (purple curve) showed a better performance compared to serum PSA (black curve, all grades AUC = 0.6020; high-grade PCa AUC = 0.5690).Multiple logistic regression analysis for the combination of biomarker levels (quantification by ELISA) with the patient’s age. (A) Pearson correlation matrix showing the correlation coefficients of the seven biomarkers, age and serum PSA with each other. A correlation between variables is defined as low for values up to ±0.3, medium for values up to ±0.5 and large for values up to ±1. (B) Combinatory analysis of immunoassay validation for the detection of healthy men. The combination of PEDF and FCER2 resulted as best pair from mass spectrometry and, in addition to age, achieved a final AUC of 0.8022 and a 39.1% specificity at 100% sensitivity. ELISA results revealed that, with an AUC of 0.8196 and a specificity of 52.2%, the best performing combination of biomarker was KRT13, FCER2, and age. Combined biomarkers showed a better performance compared to the single candidates and to serum PSA (black curve, AUC = 0.6020). (C) The combination of biomarkers with age can predict the presence of high-grade PCa. PEDF, FCER2, and age achieved a final AUC of 0.7523 and a 44.5% specificity at 100% sensitivity. By combining KRT13, FCER2, and age the performance reached an AUC of 0.7801 and a specificity of 48.1% (serum PSA is represented by the black curve, AUC = 0.5690).Immunohistochemical analysis of biomarker expression in benign and malignant prostate tissue. Overview (10× objective) of three biomarkers, which showed expression in basal cells, including respective magnifications (insets, 20× objective). (A) Positivity of KRT13 in basal cells of benign tissue, whereas (B) acinar adenocarcinoma shows loss of basal cells and KRT13 expression. (C) HPX showed in addition to expression in basal cells, reactivity in luminal cells of benign tissue, as well as obvious positivity in the fibromuscular stromal cells (background). (D) Prostate adenocarcinoma in comparison showed decreased expression of HPX. (E) PEDF showed reactivity in some of the basal cells, and weaker reactivity in luminal cells of the benign tissue. (F) In comparison, equally low expression in the (luminal) cells of the adenocarcinoma complexes.Characteristics of the patients. Statistical analysis was performed using a Mann–Whitney U test, which showed age as the only variable significantly different between the tumor vs. non-tumor patients (age: p = 0.048; PSA: p = 0.323; prostate volume: p = 0.164). * Data available for only 41 patients.ROC curve and multiple logistic regression analysis of the mass spectrometry results. The analysis was performed on the seven biomarker candidates and their possible non-correlating combinations for the identification of healthy men.ROC analysis of the ELISA results for the detection of healthy men and high-grade PCa. The table shows the diagnostic performance of ELISA results obtained normalizing the concentration of the seven candidates with two control molecules (CD44 and RNASE2). The “all PCa grades” analysis identifies healthy men (reaching 100% sensitivity at a specific threshold), whereas the “high-grade (GS 7–9) PCa” analysis identifies true negatives as either healthy men or patients harboring GS 6 PCa (reaching 100% sensitivity at a specific threshold).ROC curve and multiple logistic regression analysis of the ELISA results for the detection of healthy men or high-grade PCa. The seven single biomarkers (not normalized) and their combinations (including patients’ age as variable) were analyzed. The “all PCa grades” analysis identifies healthy men (reaching 100% sensitivity at a specific threshold), whereas the “high-grade (GS 7–9) PCa” analysis identifies true negatives as either healthy men or patients harboring GS 6 PCa (reaching 100% sensitivity at a specific threshold).Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01136.txt

@@ -0,0 +1 @@
+Liver transplantation (LT) is the only definitive treatment to cure hepatocellular carcinoma (HCC) in cirrhosis. Unfortunately, the shortage of donor livers limits access to the best available therapy. As a consequence, transplant candidates undergo selection for transparent waiting list acceptance and priority using the model for end-stage liver disease (MELD). Allocation of standard exception (SE) points for HCC inside the Milan-criteria balances the underrepresented urgency by labMELD in these patients who are exposed to tumor progression. Moreover, patients with HCC outside Milan can undergo LT without SE and might benefit from extended criteria donor (ECD)-grafts. We hypothesized that LT for Milan-out patients is associated with a more complicated postoperative course, reflecting higher costs. Interestingly, we found that LT for patients with Milan-in and Milan-out had comparable donor risk index, clinical outcome and cost-effectiveness. In conclusion, LT with ECD-grafts can have the maximum benefit for selected patients and bear limited financial risk.Liver transplantation (LT) is the only definitive treatment to cure hepatocellular carcinoma (HCC) in cirrhosis. Waiting-list candidates are selected by the model for end-stage liver disease (MELD). However, many indications are not sufficiently represented by labMELD. For HCC, patients are selected by Milan-criteria: Milan-in qualifies for standard exception (SE) and better organ access on the waiting list; while Milan-out patients are restricted to labMELD and might benefit from extended criteria donor (ECD)-grafts. We analyzed a cohort of 102 patients (2011–2020). Patients with labMELD (no SE, Milan-out, n = 56) and matchMELD (SE-HCC, Milan-in, n = 46) were compared. The median overall survival was not significantly different (p = 0.759). No difference was found in time on the waiting list (p = 0.881), donor risk index (p = 0.697) or median costs (p = 0.204, EUR 43,500 (EUR 17,800–185,000) for labMELD and EUR 30,300 (EUR 17,200–395,900) for matchMELD). Costs were triggered by a cut-off labMELD of 12 points. Overall, the deficit increased by EUR 580 per labMELD point. Cost drivers were re-operation (p < 0.001), infection with multiresistant germs (p = 0.020), dialysis (p = 0.017), operation time (p = 0.012) and transfusions (p < 0.001). In conclusion, this study demonstrates that LT for HCC is successful and cost-effective in low labMELD patients independent of Milan-criteria. Therefore, ECD-grafts are favorized in Milan-out HCC patients with low labMELD.Liver transplantation (LT) is a standard procedure in the treatment of hepatocellular carcinoma (HCC). The costs of HCC care are rising due to its increasing incidence [1]. Although LT is known to have the highest impact on quality adjusted life years (QALYs) in small tumor stages, costs for LT are also higher compared to other curative treatment options [1,2]. Nevertheless, reports on higher tumor stages are missing. In parallel with organ shortage and increasing costs of transplantation programs [3], the limits of the transplant indication are being expanded without taking financial resources into account. This also applies to HCC.Organs are allocated depending on an individual patient risk (sickest first principle) by the Model for End-stage Liver Disease (MELD) system [4,5], which was introduced in Germany in 2006 [6]. A bonus point-reporting system by the German Medical Association, comparable to the standard exception (SE) by the United Network for Organ Sharing (UNOS), exists for patients who experience an inadequate representation of the severity of the disease with the laboratory (lab) MELD. Patients with a SE-HCC, the so-called matchMELD, receive 22 points. This corresponds to a 3-month probability of dying in labMELD patients of 15%. The MELD score is then upgraded at 3-month intervals, which corresponds to an increase of additional 10% 3-month probability of dying in each case. MatchMELD criteria are fulfilled for patients within the Milan criteria (MC-in) due to their very good prognosis [7]: MC includes a single tumor lesion ≤ 5 cm in size or up to three lesions ≥ 1 and ≤3 cm [8]. However, only tumors ≥ 2 and ≤5 cm (UNOS T2) qualify for the matchMELD SE [9]. The diagnosis is usually made entirely on the basis of image morphology using the Liver Imaging Reporting and Data System (LI-RADS) criteria [10].Beyond the Milan criteria, LT can be carried out within defined tumor stages (e.g., University California San Francisco (UCSF) score) with comparable oncological outcomes [11]. However, due to their mostly low labMELD score, these patients have an inferior urgency for organ allocation compared to matchMELD patients and compete with a variety of benign liver diseases. We postulate that HCC patients with a matchMELD receive an organ faster and at a superior quality than that of the labMELD collective. The aim of this study is to investigate whether the treatment of patients without bonus points has a negative impact on the clinical outcome and costs.In the present work, a retrospective evaluation of a patient cohort with the ICD 10 diagnosis code C22.0 (HCC) was carried out. Patients were treated in a curative approach between the years 2011 and 2020 at our University Transplant Centre. Non-HCC patients, patients with mixed CCC/HCC and patients with extrahepatic tumor growth were excluded from the study. Transplant procedures have been described in detail earlier [12].Basic data, pre- and postoperative tumor classification, clinical course and follow-up were correlated with donor data and insurance provider data. The influence of the donor organ quality was analyzed using data from the Eurotransplant (ET) register, the type of organ allocation (primary organ vs. rescue allocation) and the type of organ (full organ, split organ or living donor LT (LDLT)). Data for the ET-donor risk index (ET-DRI: age of the donor, cause of death (COD) hypoxia, cerebral infarction, cardiac death (DCD—not allowed in Germany) or others, origin regional or beyond, cold ischemia period, µGT and rescue offer were taken from the ET register. The score is calculated using the following function [13]:Costs were evaluated according to cost categories and accounts in accordance with the specifications of the German Institute for the Hospital Remuneration System (Institut für Entgeltkalkulation im Krankenhaus (InEK)) calculation manual (4th version) [14] and compared with the remuneration calculated by the InEK.Data protection law was taken into account when evaluating this study. The study was approved by the local ethics committee (D 523/21). The statistical evaluation of this work was carried out with the program SPSS® 25 (IBM®, Armonk, NY, USA). A significance level of p ≤ 0.05 was assumed to be significant. The graphical representation takes place with the GraphPad Prism software (version 9.1.2; GraphPad software, San Diego, CA, USA) or for the creation of limit value optimization curves with the program MedCalc (Statcon GmbH, Witzenhausen, Germany). The median and its range is given below. Patient survival was evaluated using a Kaplan–Meier analysis followed by a log-rank test. Descriptive data analysis (number and relative frequency in the total population in %), linear regression analysis and logistic regression analysis were carried out as further statistical methods. Between 2011 and 2020, liver transplants were performed in 112 patients at our University Transplant Centre as part of the diagnosis of HCC. Due to a lack of insurance provider data for patients who were still in inpatient treatment on 31 December 2020, three patients were excluded. Seven patients with an incidental tumor finding in the explanted organ were also excluded. A total of 102 patients were included in the statistical analysis of this work. Table 1 gives details of the patient characteristics. The HCC diagnosis was based on image morphology except for biopsy in 19 cases (18.6%) before ET listing to rule out a CCC. Liver cirrhosis is the most common morphological expression of liver-cell damage as the cause of the development of HCC (n = 98; 96.1%). Leading etiologies of cirrhosis in our cohort were alcoholic liver damage (n = 42; 41.2%) and nonalcoholic steatohepatis (NASH; n = 11; 10.8%). Four noncirrhosis patients had other liver cell damage such as fibrosis. A positive AFP value was documented in 47 cases (46.1%; >8 ng/ml). Concomitant symptoms are depicted in Table 1.The listing on the ET waiting list took place after recognition of the SE criteria for HCC in 46 Milan-in patients (45.1%) with a matchMELD (Table 2). In the collective, 11 patients were transplanted as UCSF-in (10.8%). The time on the waiting list did not differ significantly between both groups (p = 0.881). As expected, tumor size and number of tumor lesions differed significantly within the groups (p = 0.005 and p = 0.004, respectively).Only 2 patients (2.0%) had dialysis before LT and 36 patients (35.3%) received abdominal surgery, of which 16 patients received a liver resection (15.7%). Locoregional therapy methods were performed in the majority of the patients, with transarterial chemoembolization (TACE) being the predominant procedure in 65 patients (63.7%) and together with selective internal radiation therapy (SIRT) in 4 (3.9%), percutaneous ethanol injection (PEI) in 2 (2.0%) and radiofrequency ablation in 3 (2.9%) patients, respectively. One patient (1.0%) received isolated PEI. In detail, in the matchMELD cohort, 39 patients (84.8%) received TACE (between 1 to 9 sessions, median 2). In sequence with TACE therapy, two patients additionally received PEI (4.3%) and two patients additionally received radiofrequency ablation (4.3%). In the labMELD cohort, 35 patients (62.5%) received TACE therapy (between 1 and 23 sessions, median 4); in 4 patients with additional SIRT (7.1%) and in 1 patient with radiofrequency ablation (1.8%). One patient received one PEI therapy without other locoregional treatment procedures (1.8%). Of note, the patient with 23 TACE sessions was on the waiting list for 4.3 years. No pharmacological HCC treatment was given before transplantation in our cohort.While 16 patients were allocated a primary offer via ET, for 86 patients (84.3%) a rescue offer was accepted. Acute liver failure and HU listing did not occur. Four patients received an organ after LDLT (3.9%), two patients received a split organ and two further patients received organs after domino transplantation (each 2.0%). The median DRI was for the entire collective as well as for the labMELD and matchMELD collective 2.27 (1.14–4.10), 2.29 (1.40–4.10) and 2.17 (1.14–3.09) and the donor age was 62 (6–86), 63 (10–86) and 61 (6–86), respectively. In total, 19 matchMELD patients (33.9%) and 22 labMELD patients (47.8%) received DDLT with a donor age ≥ 65 years. In histopathological examinations, between 0 and 25 foci were found (median 2), the diameter of the largest lesion was 14.0 cm (median 2.5). No tumor could be detected in four patients after locoregional therapy. The distribution of tumor sizes between the labMELD and matchMELD cohorts also differed significantly (p = 0.011). The largest tumor in the labMELD cohort was 14.0 cm and in the matchMELD cohort 9.1 cm. The medians in the groups were 3.0 cm and 2.4 cm, respectively. The median waiting time for a donor organ from the time of ET listing was 109 days (1–1556 days). After a previous resection (n = 16), a transplant was carried out 501.5 days later (20–1367 days). The median surgery time was 269.5 (185–497) minutes. The cold ischemia time (CIT) was 565 (135–1025) minutes for the entire collective. Antiviral therapy with the hepatitis B immunoglobulin (HBIG) was carried out in 11 patients (10.1%) due to a replicative hepatitis B infection or for hepatitis B positive donor organ (n = 4). Subsequent hepatitis B prophylaxis was given to 22 patients (21.6%). CMV prophylaxis was applied to 38 patients (37.3%). The median postoperative length of stay was 16 days (1–203 days). High urgency retransplantation were performed in the labMELD and the matchMELD cohort in 2 and 3 cases (p = 0.405).Importantly, we did not find significant differences in the DRI as well as the donor age between the labMELD and the matchMELD groups (p = 0.697 and p = 0.905).The median follow-up of the patients was 3.25 years (0–10.00 years). Of 34 deceased patients (33.3%, median 1.27 years after transplantation, range 0–7.58 years), 10 patients (9.2%) died during their inpatient stay (median 12 days after surgery, range 0–202 days). As expected, a survival analysis broken down into preoperative Milan and UCSF criteria showed no significant difference (p = 0.193, Figure 1a), although patients with UCSF-in tumors and patients outside Milan and UCSF seem to have a lower survival rate. The hypothesis was then checked whether there was a difference in survival depending on the MELD listing. The survival analysis (Figure 1b) of patients who were assigned an organ with a labMELD and patients with a matchMELD showed no statistically significant difference (p = 0.759). This result did not change when looking at the survival after diagnosis. Of note, the current follow-up of the patients with an organ age of 86 years were 3 and 10 years.The cost bearer data for the transplant stay was evaluated next (Figure 2). The median costs for a liver transplant were EUR 37,332 (EUR 17,224–395,944) with a calculated revenue of EUR 41,897 (EUR 31,586–383,020). This results in a median revenue of EUR 7725 per case. In the labMELD and matchMELD collectives, costs were at EUR 43,513 (EUR 17,837–185,011) and EUR 30,324 (EUR 17,224–395,944), revenue was at EUR 44,724 (EUR 32,078–206,717) and EUR 39,908 (EUR 31,586–383,020) and deficit was at EUR 6433 (EUR −106,149–47,876) and EUR 10,182 (EUR −33,362–40,482). All results did not differ significantly. In the present HCC cohort, donor criteria such as donor age and DRI had no influence on the reimbursement of the inpatient case (p = 0.917 and p = 0.126) or patient survival (p = 0.835 and p = 0.726).In order to evaluate a potential influence of bridging treatment on cost-effectiveness between labMELD and matchMELD groups, we analyzed costs, revenue and deficit of TACE treatment procedures, as this remarks the predominant locoregional bridging treatment. For the labMELD and matchMELD cohorts, costs were at a median of EUR 9074 (EUR 1780–53,801) and EUR 5325 (EUR 1658–28,210), remuneration at EUR 11,440 (EUR 2151–68,745) and EUR 6089 (EUR 1786–32,391) and deficit at EUR 2489 (EUR −2120–14,944) and 1223 (EUR −1113–5193), respectively. While the costs and remuneration differed significantly between the labMELD and matchMELD cohorts (p = 0.041 and p = 0.034), the deficit did not differ significantly (p = 0.112).For preoperative prediction of a deficient hospital stay, we performed linear and logistic regression analysis. In the multivariate linear regression, maximum labMELD (p = 0.037) and the labMELD at the time of transplantation (p = 0.044) were highly significant. The deficit increased by EUR 576 per increase in the maximum labMELD or by 580 EUR per increase for the labMELD at the time of transplantation. A cut-off analysis for the labMELD score was carried out using a receiver operating characteristics (ROC) curve. Figure 3 shows the ROC curves for the maximum measured labMELD (a) and the labMELD at the time of the transplantation (b). The Youden index is a MELD score of >13 points for the maximum labMELD and >12 points for the labMELD at the time of transplantation. Both results are significant at p < 0.001.In the multivariate logistic regression, a previous hepatitis C disease (p = 0.031) and a hepato-renal syndrome as a manifestation of the liver disease (p = 0.030) were found to be significantly associated with a deficit remuneration. However, in particular, the DRI, matchMELD and the waiting time on the ET waiting list did not show any influence on the costs and reimbursement of a LT. Despite numerous postoperative predictors in univariate analysis, the multivariate logistic regression showed only the need for a new operation (p < 0.001), the reinfection with multiresistant germs (p = 0.020) and dialysis (p = 0.017) to be significantly associated with a deficit. The multivariate linear regression showed the duration of the transplant operation (p = 0.012) as well as the need of erythrocyte concentrates (p < 0.001) and fresh frozen plasma concentrates (p = 0.025) significantly associated with a deficit. The financial deficit increased by EUR 86.50 per minute of surgery and by EUR 1300 and EUR 680 per erythrocyte or fresh frozen plasma concentrate. To check the quality of the parameter “operation time” as a criterion for assessing a deficit case, a ROC curve was created (Figure 4). The Youden index is an operation duration of >273 min (p < 0.001).Costs of HCC treatment are rising [1]. Under certain conditions, HCC can be treated curatively with a liver transplant. While LTs are known to have the highest impact on QALY, it marks also the most expensive treatment option [2]. So far, only HCC treatment of small tumor sizes is evaluated concerning its cost-effectiveness [2,15,16,17]. However, since patients with tumor stages larger than the Milan criteria, such as within the UCSF criteria, were demonstrated to have comparable oncological survival after transplantation [11], its burden on the health care system should be considered. We therefore intended to compare the effect of LTs for patients within and beyond Milan criteria on its cost-effectiveness. While Milan-in patients are entitled to a matchMELD, patients with a larger tumor do not receive any additional points, so that their urgency on the ET list is lower. Differences in outcome concerning this regard were discussed [11], but the financial aspect was not focused so far.We were able to demonstrate a comparable outcome within our cohort. Although we presented, as expected, a significant difference in tumor size and number of lesions between the labMELD and the matchMELD cohorts (p = 0.005 and p = 0.004, respectively), overall survival after transplantation were comparable between both groups (p = 0.759). Of note, cohorts divided between Milan-in, UCSF-in and even tumors beyond UCSF did not show a significant difference in overall survival (p = 0.193). Moreover, we did not find differences in the waiting time for a donor liver: the time of waiting on the ET list did not differ (p = 0.881). Contrary to our expectations, financial differences did not vary significantly (p = 0.204 and p = 0.492). However, earnings for the hospital were slightly higher with matchMELD patients. Interestingly, the median financial benefit in TACE treatment, which is the most commonly used locoregional treatment modality, was positive in both groups. As expected, the costs for TACE treatment were significantly higher in the labMELD group (p = 0.041). However, the remuneration was even more intensified in the labMELD group (p = 0.034). In contrast to our actual expectations, the deficit did not differ significantly between the labMELD and matchMELD groups (p = 0.112) and neither group resulted in a loss of the care provider. In general, the liver transplant program in HCC can be performed cost-effectively. In contrast, unpublished results of our working group show costs of EUR 82,569 ± 81,820 with a deficit of EUR 281 per transplant case for the entire transplant collective (2011–2016; n = 179), regardless of the primary diagnosis. In order to understand which characteristics distinguish patients from cost-effective to cost-intensive cases, we performed single and multivariate regression analysis. The maximum labMELD (p = 0.037) and for the labMELD value at the time of transplantation (p = 0.044) were identified as possible parameters to predict loss-making cases: by means of a ROC analysis, threshold values of >13 points for the maximum labMELD as well as >12 points for the labMELD at the time of operation were identified (each p < 0.001). For every increased MELD point, the deficit increases by EUR 580. This is in agreement with the results of other studies [18,19,20,21]. In an American cohort between 2002 and 2005, a comparable increase in costs per MELD point of USD 580 was demonstrated [20].In our study, the clinical condition of the patient before transplantation was only significantly associated with a deficit in compensation with regard to hepatitis C as an underlying disease (p = 0.031) and the presence of a hepato-renal syndrome (p = 0.030). Of note, donor criteria such as donor age and DRI did not have an influence on the reimbursement of the inpatient case (p = 0.917 and p = 0.126) or patient survival (p = 0.835 and p = 0.726). However, a corresponding prediction of high costs depending on the DRI score was demonstrated for heterogeneous patient collectives [22,23,24].Postoperative complications, such as peritonitis, are other factors that can increase case costs [20]. Using multivariate logistic regression, significant correlations with a deficit case were found for the need of further operations (p < 0.001), infection with multiresistant germs (p = 0.020), dialysis (p = 0.017) and, in the multivariate linear regression, the duration of the transplant operation (p = 0.012) as well as the need of erythrocyte concentrates and fresh frozen plasma concentrates (p < 0.001 and p = 0.025). In ROC analysis, an operation duration of 273 min was identified as a critical threshold value, whereby the financial deficit per operation minute increased by EUR 86.50. The need of blood concentrates (erythrocytes p < 0.001 and fresh frozen plasma concentrates p = 0.025) increased the deficit by EUR 1300 and EUR 690. Other postoperative complications did not provide any indication of the development of a cost-ineffective case. In a comparison of liver resection and LT, the literature shows a longer patient survival in favor of transplantation, although the costs are significantly higher [15]. A positive cost–benefit assessment was shown for patients with survival after transplantation > 2 years [15]. This supports the need for a preoperative assessment of survival after LT on the basis of demographic, clinical and donor-specific data. So far, in different calculation models, prognoses of the financial outcome of a LT in comparison to resection were performed, although, they only consider small tumors in Child-Pugh A and B cirrhosis [16,17].This study has several limitations. First of all, it is a unicentric, retrospectively collected data query in which the results can differ due to variables that are not considered in the regression analysis. The results should be compared with those of other transplant centers, although the data protection principles for this are lacking. Due to the small number of cases of n = 16 resections before subsequent transplantation, a comparison between resection and transplantation is not possible on the basis of the available data and requires a larger cohort. In addition, financial analyses comparing RFA, TACE, resection and transplantation, especially in the case of tumor manifestations outside the Milan criteria, should be tackled.Patients with HCC outside Milan benefit from LT and show comparable survival to MC patients without financial burden for the caregiver. Our HCC transplant program is cost-effective for the inpatient stay as well as for the locoregional bridging treatment with TACE. It can be stated that there is no economic effect in terms of awarding bonus points for the HCC. Regardless of monetary pressure, patients outside the MC can undergo LT depending on the medical indication. The use of ECD organs is highly effective and not a risk for either the patient or the hospital. The survival of our cohort with a follow-up of 3.85 years is 67.7% and the hospital mortality is low, showing 9.2%, which is within the upper limit of 20% tolerated by the German IQTIG (Institute for quality assurance and transparency in health care). In addition, due to scarce resources and steadily increasing health expenditure, it seems appropriate to consider financial aspects in order to reveal wrong financially motivated triggers.Conceptualization, J.-P.G., T.B. and F.B.; formal analysis, J.-P.G.; investigation, J.-P.G. and F.W.; resources, T.B.; writing—original draft preparation, J.-P.G.; writing—review and editing, J.-P.G., M.L., H.D. and F.B.; visualization, J.-P.G.; supervision, F.B.; project administration, T.B. and F.B. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of the Medical Faculty of the Christian-Albrechts-University of Kiel (D523/21; date of approval: 6 April 2021).Informed consent was obtained from all subjects involved in the study.The clinical datasets supporting the conclusions of this study were derived from patient files (paper and electronic form). Therefore, restrictions to availability apply due to data protection regulation. Anonymized data are, however, available from the corresponding author on reasonable request and with permission of the University Hospital Schleswig–Holstein and the local review board.The authors declare no conflict of interest.Kaplan–Meier analysis for overall survival in dependence of: (a) preoperative tumor criteria; and (b) labMELD and matchMELD listing.Cost analysis: Presented are (a) costs; (b) remuneration and corresponding (c) deficit for the labMELD and matchMELD cohorts in EUR (€), respectively. All differences are not significant.Critical value analysis for the labMELD. ROC-curve for (a): maximum measured labMELD and (b): the labMELD at the time of the transplantation for the detection of a cost-ineffective hospital stay. The Youden index is a MELD score of >13 points for the maximum labMELD and >12 points for the labMELD at the time of transplantation.Critical value of the surgery time in ROC analysis.Patient characteristics.Abbr.: TIPS—transjugular intrahepatic portosystemic shunt. p-values from univariate regression are displayed.Transplant criteria & MELD score.Represented are number (n) and percentage (%) of the cohorts as well as median (range) for waiting time, diagnostic tumor size and MELD scores at liver transplantation (LT). For retransplantation, number of retransplantation within hospital stay of initial transplantation and number of retransplantation during follow-up are presented. All retransplantations within initial stay were high urge within 2 weeks. p-values from univariate regression are displayed.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01137.txt

@@ -0,0 +1 @@
+Small bowel adenocarcinomas (SBA) are rare tumors with a poor prognosis. Due to the rareness of this illness, there is limited scientific data to guide therapeutic management. The recent large prospective ARCAD-NADEGE study evaluated 347 patients with SBA and has helped to improve our current knowledge of this disease. When diagnosed with advanced, non-surgically resectable disease, chemotherapy remains the cornerstone of the treatment and appears to add a survival benefit compared to palliative care. Other more recent drugs, such as small molecule targeted therapeutic inhibitors or immunotherapy, may have a role in improving the outcome of patients with advanced SBA. In this review, we summarized the classical therapeutic arsenal as well as possible future promising treatments in advanced SBA.Small bowel adenocarcinoma (SBA) is diagnosed at an advanced (unresectable or metastatic) tumor stage in approximately one-third of cases. This is partly due to the non-specific symptomatology and limitations in endoscopic and radiologic detection methods. In this context, the prognosis remains poor and systemic chemotherapy appears to benefit patients when compared to best supportive care alone, despite the absence of randomized controlled trials. The results of a recent large prospective cohort (ARCAD-NADEGE) reported that the absence of chemotherapy was a predictive factor for a lower overall survival (OS) even though poor differentiation and SBA associated with Crohn’s disease correlate with poor prognosis. In retrospective series, the median OS ranges from approximately 9 to 18 months with current treatment approaches. A combination of a fluoropyrimidine and oxaliplatin (FOLFOX or CAPOX) appears to be the most utilized and effective first-line chemotherapy regimen. Other front-line alternatives are the combination of 5-FU and cisplatin or fluoropyrimidine and irinotecan (FOLFIRI). In second-line, FOLFIRI is an effective option after progression on platinum-based therapy. Taxane-based therapy appears to be an alternative option, but further evaluation in larger series is needed. To a limited extent, the role of surgical resection for metastatic disease appears to be a valid option, though this approach has not been evaluated in prospective clinical studies. Due to the rareness of the disease, inclusion in clinical trials should be prioritized, and there is hope that targeted therapies and immunotherapy may enter the therapeutic arsenal for these patients. Small bowel adenocarcinoma (SBA), duodenal, jejunal, or ileum localizations, is a rare disease and accounts for 3–5% of gastrointestinal cancers. These tumors are diagnosed at an advanced (unresectable or metastatic) stage in around one-third of cases, in part due to the non-specific nature of clinical symptoms. With the development of improved imaging techniques and advances in endoscopy, such as enteroscopy and capsule endoscopy, it is possible that trends toward earlier diagnosis can be seen in the future [1,2,3]. In the NADEGE cohort study, the tumor stage at diagnosis was localized or resectable in 54%, locally advanced, unresectable in 5.5%, metastatic in 33.5%, and undetermined clinical stage in 7% [4]. As in many other cancers, the most important prognostic factor in SBA is the TNM classification. Despite a relative improvement in the median overall survival (OS) for localized disease, the overall prognosis for SBA remains poor. The 5-year overall survival (OS) rate is 50% for stage I, 40% for stage II, 10–40% for stage III, and reaching less than 5% for stage IV disease [5]. The other factors associated with a worse prognosis are male gender, duodenal location, poor differentiation, and SBA associated with Crohn’s disease as compared to de novo SBA [5,6,7,8]. Retrospective studies demonstrated the ability to resect limited metastatic disease, but no prospective studies were conducted. For unresectable metastatic disease, resection of the primary tumor should only be considered in the case of primary tumor symptomatology, such as perforation, bowel obstruction, or uncontrolled gastrointestinal bleeding. For all other situations, the main modality of therapy for metastatic SBA is palliative chemotherapy. This review focuses upon the role of systemic therapy, surgical metastasectomy, and the novel therapies such as targeted therapies and immunotherapy in advanced SBA.Although no randomized clinical trials have yet reported a benefit for systemic chemotherapy compared to best supportive care alone, retrospective studies suggested a benefit for palliative chemotherapy. One of the earliest retrospective studies reported overall survival (OS) of 12 versus 2 months for systemic chemotherapy and best supportive care, respectively [6]. Several retrospective or prospective studies since reported a survival benefit for the administration of chemotherapy compared to best supportive care alone, with a median OS ranging from 9 to 19 months in patients with palliative chemotherapy, whereas it ranged only from 2 to 13 months in patients with palliative care [4,6,9,10,11,12,13,14] (Table 1).The results of a recent large prospective cohort (NADEGE) reported that the only predictive factor of a lower OS was the absence of chemotherapy. These data are, however, subjected to bias as patients who did not receive chemotherapy were more likely to be frail and have a poorer performance status [4]. In this observational study, a benefit of palliative chemotherapy was described with a median OS of 14.3 months for the 86 patients receiving systemic chemotherapy versus 2.2 months for 15 patients treated with palliative care only. Among all patients with metastatic disease, the median OS was 12.7 months [4]. Furthermore, although having a prognostic impact, the tumor location had no impact on the response to chemotherapy.These survival results are consistent with those reported in several phase II studies with an OS ranging from 8 to 20 months and a PFS ranging from 3 to 11 months for patients treated with various chemotherapy regimens [15,16,17,18,19,20] (Table 2).In most studies: first-line chemotherapy was mainly 5-FU based regimens, with or without a platinum salt or irinotecan. These different regimens are shown in Table 1 and Table 2 [15,16,17,18,20,21,22,23,24,26,27,28].In the NADEGE cohort, chemotherapy was given to 122 metastatic SBA patients with regimes being FOLFOX or capecitabine and oxaliplatin (CAPOX) in 69.8% of patients, FOLFIRI in 18.6% of patients, and 5-FU monotherapy in 7.0% of patients [4]. In summary, a fluoropyrimidine and oxaliplatin (FOLFOX or CAPOX) appear to be the most used and effective first-line chemotherapy.One of the first studies of chemotherapy in SBA involving eight advanced SBA patients reported a median OS of 13 months and a median PFS of 7.8 months for patients treated with 5-FU-based chemotherapy. A second early study of 20 patients reported a median OS of 14 months for patients treated with the combination of 5-FU and a platinum salt [21,22].The benefit of 5-FU combined with a platinum salt in comparison to other chemotherapy regimens is suggested by one retrospective study involving 80 patients with advanced SBA. In this series, median PFS and OS were 8.7 and 14.8 months, respectively, for 5-FU and cisplatin, which was significantly better than other chemotherapy regimens [23].In a retrospective Japanese single-center study involving 132 patients, Tsushima et al. reported that FOLFOX demonstrated improved PFS and a trend towards better OS when compared to other chemotherapy regimens [26]. The potential benefit for a fluoropyrimidine and a platinum agent over other chemotherapy is supported by a retrospective multicenter study reported by Zaanan et al. In this study involving 93 SBA patients, FOLFOX treated patients had an improved OS of 17.8 months in comparison to other chemotherapy regimens such as 5-FU alone or combined with cisplatin or irinotecan [24]. In the subgroup of patients treated with platinum-based chemotherapy, multivariate analysis demonstrated that the FOLFOX regimen in comparison to the combination of 5-FU and cisplatin was associated with an improved median PFS, 6.9 versus 4.8 months (p < 0.0001), and an improved median OS, 17.8 versus 9.3 months (p = 0.02), respectively. For all first-line regimens, the median OS was 15.1 months [24].These results are similar to those from three prospective phase II trials, in which encouraging objective response rate (ORR) and OS were reported for the combination of a fluoropyrimidine and oxaliplatin (FOLFOX or CAPOX) in the front-line setting. In these three trials, ORR ranged from 45 to 50%, and median OS ranged from 15.2 to 20.4 months [16,17,18] (Table 2). In these studies, toxicities were generally mild, with the most common grade III or IV treatment-related toxicities being hematological (neutropenia or thrombocytopenia) in 10% to 63%, fatigue in 3% to 30%, diarrhea in 3% to 10%, nausea or vomiting in 3% to 10%, and peripheral neuropathy in 9% to 25% of the patients.The association of 5-FU, mitomycin, and doxorubicin was evaluated in one phase II multicenter study involving 39 patients with SBA or adenocarcinoma of the ampulla of Vater. This combination demonstrated disappointing results with an ORR of 18.4% and a median OS of 8 months [15]. As expected, the most frequent toxicities were hematological and gastrointestinal, with grade III to V toxicities reported in 26 (72%) of the 36 patients evaluated and with one patient dying from grade V hematological toxicity [15].Another first-line option is triplet chemotherapy consisting of 5-FU, oxaliplatin, and irinotecan (CAPIRINOX or FOLFIRINOX) as it is utilized for metastatic colorectal or pancreatic cancer. However, in a recent phase II study involving 33 patients, McWilliams et al. reported a median OS of 13.4 months and a disease control rate of 37.5% in patients with advanced SBA with the first-line use of CAPIRINOX [20]. Although not formally compared, the response rate and survival for CAPIRINOX are lower than those described in the phase II trials with CAPOX or FOLFOX [16,17,18,20] (Table 2). Few data exist for second-line chemotherapy (Table 2). In the small series by Locher et al., second-line FOLFIRI demonstrated a median PFS of 5 months in eight patients. Among these eight patients, five had a clinical benefit [22]. These results are supported by a multicenter retrospective series reported by Zaanan et al., in which 28 patients treated with FOLFIRI demonstrated an ORR and disease control rate of 20% and 50%, respectively. In the same study, median PFS and OS were 3.5 months and 10.5 months, respectively [25]. In this series, 48% of the patients had grade III or IV therapy-related adverse events, mainly neutropenia in 37% and gastrointestinal in 14%. One treatment-related death due to grade V neutropenia was reported [25]. These data suggest a modest activity and a relatively manageable toxicity profile for FOLFIRI as a second-line treatment in patients with progressive advanced SBA after platinum-based first-line chemotherapy. Two other small studies suggested clinical activity for taxane-based therapy in advanced SBA after first-line chemotherapy [19,28]. In a retrospective study, Aldrich et al. reported the results from taxane-based therapy in 20 patients demonstrating an ORR of 65% and a median PFS and OS of 3.8 and 10.7 months, respectively [28]. A prospective phase II study reported by Overman et al. reported similar results for 10 evaluable patients with advanced SBA treated with nab-paclitaxel with median PFS and OS of 3.2 and 10.9 months, respectively [19]. Although the sample sizes are small, taxane chemotherapy might represent a novel therapeutic option for SBA patients; however, replication in larger datasets is needed [19,28].Limited data exist regarding other chemotherapy regimens in advanced SBA. Some responses with gemcitabine were described in several studies, suggesting a benefit from this drug [9,23]. In the retrospective series by Fishman et al., an ORR of 33.3% for gemcitabine alone and an ORR of 50% for the combination of a fluoropyrimidine and gemcitabine in first or second-line setting was reported [8]. Another study reported no responses in the first-line therapy with gemcitabine but one response among two patients treated by gemcitabine in the second-line setting [23]. More recently, a retrospective study reported by Aydin et al. demonstrated an ORR of 20%, a median PFS of 6 months, and a median OS of 11 months among 10 patients treated by gemcitabine monotherapy in the first-line setting [27]. These data appear to support the clinical activity of gemcitabine in advanced SBA. Due to some similarities between SBA and colorectal cancer (CRC) in terms of molecular alterations and effective chemotherapy regimens, targeted therapies usually used in CRC, such as anti-epidermal growth factor receptor (EGFR) or anti-vascular endothelial growth factor (VEGF), were studied in advanced SBA. The role of anti-angiogenic therapies, such as bevacizumab or ramucirumab, which were well established in metastatic CRC, underwent limited evaluation in advanced SBA, despite the known importance of the VEGF pathway in SBA [29]. Indeed, immunohistochemical expression of VEGF-A was recently reported as a potentially useful biomarker for the prediction of the efficacy of bevacizumab-based treatment in patients with advanced SBA [30].Most studies evaluating bevacizumab in advanced SBA are retrospective and have suggested a survival benefit with the addition of bevacizumab to standard chemotherapy (Table 3) [31,32,33].Only one phase II trial evaluated the combination of bevacizumab and CAPOX in metastatic SBA or ampullary adenocarcinoma. This study reported a 6 month PFS rate of 68% and an ORR of 48% [34]. However, despite the absence of statistical comparison, these results did not appear improved compared to similar results reported from an earlier conducted phase II trial of CAPOX [16].An ongoing randomized phase II trial is currently evaluating ramucirumab and paclitaxel versus FOLFIRI in refractory SBA previously treated by a fluoropyrimidine and/or oxaliplatin. This trial may help define the therapeutic algorithm for second-line chemotherapy in advanced SBA (NCT 04205968) (Table 3).By analogy with CRC, studies have evaluated the benefit of anti-EGFR antibodies in RAS wild-type advanced SBA [35,36,37,38] (Table 3). Indeed, the proportion of RAS mutations in SBA is around 40% to 53%, initially suggesting a theoretical potential efficiency of anti-EGFR antibodies may apply to approximately one-half of all SBA [43,44].An early case series of four patients suggested a benefit from cetuximab when combined with irinotecan in patients with advanced SBA. Among these four patients, one patient obtained a complete response, two patients had a partial response, and one patient had stable disease at the first evaluation. Among the three patients tested for KRAS mutations, all had a wild-type mutational status and corresponded to the two partial responses and the one complete response [35]. Another retrospective multicenter study involving 13 patients with metastatic SBA receiving anti-EGFR antibody (cetuximab or panitumumab) in monotherapy or in association with chemotherapy in first- or second-line treatment reported a median PFS of 5.5 months, a median OS of 15.8 months, and a complete response rate of 15%, a partial response rate of 39%, stable disease rate of 23%, and progression disease rate of 15%. However, in this work, RAS mutational status was not reported [36].By focusing on KRAS wild-type metastatic SBA patients, another retrospective single-center study reported on 25 patients treated with either cetuximab (n = 19) or panitumumab (n = 6) as a single agent or in combination with chemotherapy. In this study, an ORR of 12%, a DCR of 36%, and a progressive disease rate of 64% were reported [37].The results of a recent single-arm phase II study also reported disappointing results with no responses seen from panitumumab monotherapy in nine metastatic RAS wild-type SBA and ampullary adenocarcinoma patients [38].Apart from RAS mutations, EGFR mutations were reported in approximately 2.5% of SBA [44], leading to an evaluation of EGFR tyrosine kinase inhibitors, such as erlotinib in advanced SBA. In a case report of a patient with concomitant lung and duodenal adenocarcinoma with an EGFR mutation, the treatment of erlotinib and S-1 resulted in a partial response [39] (Table 3).The recently improved understanding of the mutational landscape of SBA may lead to specific targeted therapies, such as anti-HER2 antibodies, anti-MEK tyrosine kinase, PIK3CA inhibitors, or NTRK-directed therapies. The use of anti-HER2 antibodies was reported in one case report in which aHER2 amplified duodenal cancer demonstrated a response to trastuzumab and FOLFOX in the neoadjuvant setting [45].Though NTRK inhibitors have not been evaluated in advanced SBA, they already demonstrated anti-tumoral activity in a variety of tumor types with NTRK fusions [46]. It might therefore be administrated in case of this molecular alteration in advanced SBA [47].Immune checkpoint inhibitors (ICIs) appear today as the cornerstone of immunotherapy in several cancer types and are mainly represented by antibodies targeting anti-programmed cell death protein 1 (PD-1), anti-programmed cell death ligan-1 (PD-L1), and anti-cytotoxic T-lymphocyte antigen-4 (CTLA-4). Predictive biomarkers of ICIs, such as PD-L1 expression, combined positive score (CPS), microsatellite instability (MSI), and tumor mutational burden (TMB), are currently used in several types of cancers [48,49,50,51,52]. Predictive biomarkers for the efficiency of ICI in SBA remain under evaluation.The phase II basket trial, KEYNOTE-158, investigated pembrolizumab, an anti-PD-1 antibody, in advanced MSI solid tumors that experienced failure with prior therapy. For the subgroup of 19 patients with advanced SBA, the reported ORR was 42.1%, median PFS was 9.2 months, and median OS was unreached [40]. In the recent ZEBRA multicenter phase II study involving 40 patients with unresectable or metastatic SBA, pembrolizumab was evaluated as second-line treatment. In this study, 50% of the four patients with MSI-high tumors had a partial response, while 3% of the 32 MSS patients had a partial response. MSI status was unknown for four patients in this study [41].Another single-agent, open-label, phase II study evaluating avelumab, an anti-PD-L1 antibody, in patients with advanced or metastatic SBA reported that avelumab was considered safe, and antitumor activity was observed as evidenced by a 29% RR and 71% DCR [42].Combinations of ICIs together or ICIs combined with targeted therapies or chemotherapy are also being investigated. Two phase II studies are currently ongoing for the treatment of rare tumors, and both include SBA. The first trial combines ipilimumab (an anti-CTLA-4 antibody) with nivolumab (an anti-PD-1 antibody) compared to nivolumab monotherapy (NCT 02834013). The second trial compares cobimetinib (a mitogen-activated protein kinase (MEK) inhibitor) and atezolizumab (an anti-PD-L1 antibody) (NCT03108131). The results of targeted therapies of ICIs investigated in advanced SBA are summarized in Table 3.A major challenge for SBA is the need to develop biomarkers to predict clinical benefits from ICIs. Apart from biomarkers such as MSI status, tumor mutational burden, and tumor-infiltrating lymphocytes, other potential biomarkers of interest should be explored. For example, the gut microbiome could have crosstalk with cancer immune response and immunotherapy [53]. Furthermore, ferroptosis, a form of regulated cell death mainly relying on iron-mediated oxidative damage and subsequent cell membrane damage, seems to affect the efficacy of cancer treatments, and thus combinations with agents targeting ferroptosis signaling may be of relevance to SBA [54]. However, there is currently no specific data evaluating the gut microbiome or ferroptosis with regard to tumor response in the treatment of SBA.There are limited data regarding the prognosis of patients with resected metastatic SBA. Therefore, outcomes of a sub-group of 34 patients with curatively resected metastatic SBA of the ARCAD-NADEGE study were analyzed. The metastatic sites were mainly peritoneal (29.4%), liver (26.5%), lymph nodes (11.8%), lung (2.9%), multiple (14.7%), or other (14.7%) [55]. The median OS for these patients undergoing curative-intent metastatic resection of SBA was 28.6 months, which is better than the median OS of 12.7 months for all metastatic patients in the NADEGE cohort. Although there are major limitations to cross-trial comparisons, this OS compares favorably to those recently reported in studies of patients treated with palliative chemotherapies [15,16,17,18,19,20]. However, patients with resected metastasis in the ARCAD-NADEGE cohort were highly selected patients with 85% (n = 29) of patients with a solitary metastatic site and therefore not representative of all metastatic SBA patients. Among all metastatic SBA resected patients, 30 (88.2%) also received perioperative chemotherapy, though perioperative chemotherapy was not associated with a better OS in this subgroup of patients [55]. Negative predictive factors for OS after metastasectomy were poor differentiation, positive margins, and lymphatic invasion. Peritoneal carcinomatosis is a frequent site of disease ranging from around 25 to 50% of metastatic SBA [6,55,56]. Hyperthermic intraperitoneal chemotherapy (HIPEC) combined with cytoreductive surgery (CRS) was evaluated for SBA patients with peritoneal carcinomatosis. In eight observational studies, survival outcomes and toxicities were evaluated [57,58,59,60,61,62,63] (Table 4). In these studies, median OS from CRS + HIPEC ranges from 16 to 47 months, and grade III or IV treatment-related toxicities range from 12 to 35% (Table 4). The main observed complications are post-operative infections, abdominal collections, hematological toxicity, re-interventions, and pleural effusions. In the largest multicenter study by Liu et al. with 152 patients receiving CRS + HIPEC between 1989 and 2016, the median OS was 32 months with a median disease-free survival of 14 months. In the multivariate analysis, a Peritoneal Cancer Index (PCI) ≤ 15 was independently associated with an improvement in OS (p = 0.003) [62]. However, despite a tolerable rate of grade III or IV treatment-related toxicities (19.1%) there was a 2% death rate due to multiorgan failure.Thus, in SBA patients with limited peritoneal carcinomatosis (PCI ≤ 15) and a physical status allowing for a major surgical procedure, CRS and HIPEC could be considered in expert centers [57,59,60,61,62,64,65].Systemic chemotherapy appears to be benefit patients with advanced SBA compared to best supportive care alone, though data from randomized controlled trials are lacking.The combination of a fluoropyrimidine and oxaliplatin (FOLFOX or CAPOX) remains the most used and effective first-line chemotherapy with manageable toxicities. Alternatives in the first-line are 5-FU and cisplatin or FOLFIRI. In the second-line setting, FOLFIRI appears to be an effective option. Taxane-based therapy also appears to be clinically active, but further evaluation in larger series is needed. The benefit of targeted therapies for SBA is uncertain and continues to be investigated. Immune checkpoint inhibitors have demonstrated robust activity for the subset of SBA with MSI. Due to the rareness of the disease, inclusion in clinical trials should be prioritized when feasible. For selected patients, surgical resection of metastatic disease can be considered, especially in cases of isolated peritoneal, liver, or lung metastasis. For the specific location of peritoneal carcinomatosis, SBA patients with limited peritoneal disease should be considered for HIPEC and CRS at centers of excellence.Conceptualization, E.M. and A.Z.; methodology, A.Z.; software, E.M.; validation, E.M. and A.Z.; formal analysis, A.Z.; investigation, E.M.; resources, E.M.; data curation, E.M.; writing—original draft preparation, M.J.O. and E.M.; writing—review and editing, E.M.; visualization, A.Z.; supervision, A.Z.; project administration, A.Z. All authors have read and agreed to the published version of the manuscript.This study received no external funding.E.M. declares a consulting and/or advisory boards for Servier. A.Z. declares a consulting and/or advisory boards for Amgen, Lilly, Merck, Roche, Sanofi, Servier, Baxter, MSD, Pierre Fabre, Havas Life, Alira Health, Zymeworks. M.J.O. declares consulting and/or advisory boards for Takeda Pharmaceuticals, Ipsen Biopharmaceuticals, Pfizer, Merck, Glaxosmithkline, Nouscom, Gritstone, 3D Medicine, Phanes Therapeutics. The other authors declare no conflict of interest.Retrospective and prospective studies comparing chemotherapy versus best supportive care in advanced small bowel adenocarcinoma.Abbreviations: DCR: disease control rate; CT: chemotherapy; NA: not available; ORR: objective response rate; OS: overall survival * 23 patients with NA data.Evaluation of different chemotherapy regimens in advanced small bowel adenocarcinoma.Abbreviations: ORR: objective response rate; OS: overall response; PFS: progression-free survival; CT, chemotherapy.Main studies evaluating targeted therapies or immune checkpoint inhibitors in advanced small bowel adenocarcinoma.Abbreviations: CT: chemotherapy; DCR: disease control rate; MSI-H: Microsatellite instability high; NS: non-significant; ORR: objective response rate; mOS: median overall survival; mPFS: median progression-free survival; CR: complete response; PR: Partial response; SD: stable disease.Main observational studies evaluating intraperitoneal chemotherapy and cytoreductive surgery.Abbreviations: OS: overall survival; CRS: cytoreductive surgery; HIPEC: hyperthermic intraperitoneal chemotherapy.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01138.txt

@@ -0,0 +1 @@
+Cell developmental programs used in wound healing and development such as the epithelial-to-mesenchymal transition (EMT) are frequently coopted by solid tumors to increase motility, plasticity, and invasive characteristics which promote metastasis. Identifying and quantifying the presence and extent of these programs can help to aid in patient prognosis and dictate therapeutic decision making. Here, we review the methods and findings to detect and quantify these cellular transitions in both laboratory and clinical settings.The epithelial-to-mesenchymal transition (EMT) and its reversal, the mesenchymal-to-epithelial transition (MET) are critical components of the metastatic cascade in breast cancer and many other solid tumor types. Recent work has uncovered the presence of a variety of states encompassed within the EMT spectrum, each of which may play unique roles or work collectively to impact tumor progression. However, defining EMT status is not routinely carried out to determine patient prognosis or dictate therapeutic decision-making in the clinic. Identifying and quantifying the presence of various EMT states within a tumor is a critical first step to scoring patient tumors to aid in determining prognosis. Here, we review the major strides taken towards translating our understanding of EMT biology from bench to bedside. We review previously used approaches including basic immunofluorescence staining, flow cytometry, single-cell sequencing, and multiplexed tumor mapping. Future studies will benefit from the consideration of multiple methods and combinations of markers in designing a diagnostic tool for detecting and measuring EMT in patient tumors.During the progression of many solid tumors, cells at the primary tumor site undergo phenotypic changes in response to extracellular stimuli [1,2], one among these being an epithelial-to-mesenchymal transition (EMT). This embryonic developmental program increases invasive and migratory behavior that is advantageous to a metastasizing cancer cell [3], enabling them to disseminate to distant organs. Plasticity within this transition, including its reversal (mesenchymal-to-epithelial transition; MET) to regain epithelial and proliferative characteristics, has been demonstrated in metastatic colonization [4,5]. Indeed, patients whose tumors express high levels of EMT signatures have worse overall prognoses and increased rates of metastasis [6,7]. Previous works questioned the relevance of EMT in metastasis in breast [8] and pancreatic [9] cancers; however, these conclusions were based in either incomplete disruption of the intermediate EMT [10] or in lineage markers related to a highly mesenchymal state [11].Critical to our understanding of epithelial–mesenchymal plasticity (EMP) and its underlying regulators is our ability to distinguish unique EMT states from one another for identification in vitro, in vivo, and in patient samples. EMP and heterogeneity have frequently been associated with poor patient outcomes [6,7]; however, no robust method for assessing either of these has been developed to complement histopathological assessment in the clinic setting. The presence and role of a variety of hybrid EMT states in disease progression and metastasis remains a lynchpin in EMT-based therapies. Current evidence suggests that rather than relying solely on an MET to revert from a fully mesenchymal state, metastasis may result from the high plasticity and adaptability of the intermediate or hybrid states [12,13], as observed by the presence of intermediate circulating tumor cell clusters (CTCs clusters) [14,15]. Regardless of a clear-cut mechanism, which is still currently in debate, the presence of an intermediate state appears to be critical to the formation of metastasis, either through plasticity or as a transitional state.Here we review the various methods that have been utilized to identify the spectrum of E–M states within a sample, from flow cytometry to single-cell analysis of the intricate RNA and protein expression patterns found in mouse and human tumors. Ultimately, one or a combination of these methods could be applied to assess patient prognosis by providing rapid and comprehensive analysis of the EMT state and heterogeneity of tumors to inform disease aggression and treatment regimens.Several markers have been used over the years, firstly to determine the occurrence of EMT and, more recently, to distinguish various distinct states along the epithelial-to-mesenchymal spectrum. These markers are based on a range of properties, from those that inform stemness, to those that indicate morphological changes, and transcriptional regulators (EMT and MET TFs) of the transition. These markers have been used in various combinations, each with their own benefits and deficits, based on context, specificity, and ease of use (Table 1).The first discovered and readily utilized markers for EMT relate directly to the morphological changes that cells undergo to enhance invasion and motility, such as loss of classical epithelial adherens tight junction proteins and gain of non-canonical alternative intermediate filaments. E-cadherin, a key component of adherens junctions, was first identified as lost in epithelial cells that gained invasive characteristics [16]. Other adherens and tight junctional components that are key indicators of the epithelial state include claudins, occludins, and catenins as well as desmosomal components, such as desmoglein and desmocollin [17]. On the other hand, markers, such as Vimentin [18,19], fibronectin, N-cadherin, and smooth muscle actin (SMA) [20], have all been used to identify mesenchymal-like cells as invasion and/or progression markers in multiple cancer types. Co-expression of one or more epithelial markers along with mesenchymal markers, e.g., E-cadherin and Vimentin, is frequently used to identify intermediate or hybrid EMT states [21]. Morphological markers serve as good tools for defining the EMT state because they reflect the morphology of the cells themselves. However, they are frequently expressed at varying degrees across the EMT spectrum and therefore make poor singular identifiers for any individual state.While epithelial and mesenchymal markers, such as E-cadherin, vimentin, and fibronectin, serve to describe the internal cellular processes of EMT, they can be difficult to identify without permeabilization of the cell membrane, given their predominant intracellular localization. Consequently, cell surface markers and receptors have been adopted to identify and isolate E–M states while maintaining cell viability. EMT states were first stratified by CD44 and CD24 [22], and later by CD104 (ITGB4) [23], to identify tumorigenic populations of cells, linking the invasive and disease progressing nature of EMT to stem-like processes of cancer stem cells, particularly in breast cancer [24,25]. EpCAM, an epithelial cell adhesion molecule similar to E-cadherin, has been used in many ways to identify cells exhibiting an epithelial state, particularly CTCs [26], and as a marker for flow cytometry [27]. Notably, Pastushenko et al. [28] profiled a panel of cell surface markers to describe the transitions across an EMT, identifying a gating strategy using Epcam, CD106, CD51, and CD61 that accurately captured cells in an intermediate state. These will be discussed in more detail below. Cell surface markers are easy to access and utilize for cell sorting, but ultimately have proven inconsistent across tumor types and models for accurately defining EMT states.In addition to morphological characteristics, gene regulators of EMT or MET, such as transcription factors, provide a finer metric for measuring the progression of a cell across epithelial and mesenchymal transitions, and have been reviewed extensively [29]. These markers provide specificity, particularly when paired with morphological features. Master EMT regulator TFs, such as Snail [30], Twist1/2 [31], and ZEB1 [32,33], were originally identified as repressors of E-cadherin and regulators of plasticity and EMT. PRRX1 has also been implicated in later EMT staged in colorectal [34], thyroid [35], and gastric [36] cancers. Conversely, OVOL1/2 are required for the suppression of EMT and induction of MET in breast [37] and skin [38] epithelial and cancer [39]. These markers, as well as others detailed in other reviews [29] have been used extensively in transcriptomics-based approaches to rank EMT [40] as well as image-based methods, described later in this review. Transcription factors provide detailed information on EMT state, especially when analyzed as a network. However, some TFs have tissue specific functions that can convolute a global EMT signature generated in this way.A summary of key EMT marker proteins and their use and relevance in determining EMT state.While several other cytoskeletal proteins, such as FSP1 [41] and α SMA [42], secreted proteins, including fibronectin [43] and MMPs [44], and epithelial junctional proteins, such as claudins and occludins [45], have been employed as EMT markers in different contexts, these have not specifically been used to identify intermediate/hybrid EMT states and could possibly highlight cells that reside in more extreme epithelial or mesenchymal states.Immortalized or cancer derived cell lines have been used for decades as models to understand cancer at a basic level. They are easy to work with, highly manipulable, and can provide a basis for testing novel drugs and therapies. Cell lines have been used to perfect many of the methods detailed below, including flow cytometry, immunofluorescence, and RNA-sequencing. Databases, such as the cancer cell line encyclopedia (CCLE) and ATCC, serve as repositories for data and frozen stocks of cell lines for research use. However, cell lines alone cannot recapitulate the complexities of an in vivo system, which can be achieved through orthotopic transplantation in mice, rats, and other model organisms. This model, therefore, serves as a necessary but simple steppingstone to understanding E–M heterogeneity and plasticity in patients.In efforts to recapitulate human tumor progression for laboratory study, many different non-human models have been generated that mimic aspects of patient disease to study the roles of EMT and MET in tumor development, progression, and metastasis. Most popular are genetically engineered mouse models (GEMMs), although zebrafish [46,47], drosophila [48,49], and sea urchin [50,51] models have been elegantly used to generate important insights in the field. These GEMMs provide an excellent framework for studying the metastatic cascade, allowing for spontaneous tumorigenesis in a specific tissue of choice, collection of organs and circulating tumor cells for basic research and diagnostic development, and testing novel drug targets in a complex living system. GEMMs have been used with great success to isolate and study EMP in vivo in breast [28], skin [28], pancreatic [52], and prostate [53] cancer. Additionally, immune compromised mouse models can host human-derived cell lines or patient-derived xenografts, which, when transplanted orthotopically into the tissue of origin, can recapitulate the nuanced disease of that individual for further study. Overall, mouse and other models of human cancer have been crucial in expanding on in vivo dynamics of the metastatic cascade where cell lines have fallen short.Ultimately, the best tool for studying human disease is directly on human patient samples. This can be tumor or tissue biopsies either taken fresh or stored in a formalin fixative, as well as circulating tumor cells, cytological specimens, and secondary site biopsies, etc. However, acquiring patient samples and full patient data can be challenging and take years. Databases, such as The Cancer Genome Atlas (TCGA), contain complete sequenced genomes for thousands of primary patient samples and can be a helpful bioinformatic tool to begin transitioning from basic to translational research, such as validating cancer predictor genes or looking for large trends across many samples. EMP has been successfully identified, validated, and explored in patient cancers, including in circulating tumor cells [7,54,55], determining EMT gene signatures in primary tissue [56,57], and mapping EMT states at single-cell resolution [58,59], both substantiating its relevance in patient disease and opening new possibilities for diagnostic approaches.Circulating tumor cells in the blood have served as a “window to cancer” for many years [60,61]. As a diagnostic tool, it is easy to implement on patients, requiring only a small blood sample, and can be used to screen tumor cells in a multitude of ways [62], including immunofluorescence [54], RNA in situ hybridization (ISH) [15], and RNA-sequencing [14]. CTC collection methods have also helped to validate the significance of an EMT in the metastatic cascade [15,54], as well as a possible requirement for the reversal, or MET, to colonize metastatic organs [63].Traditionally, circulating tumor cells were harvested using the cell surface marker EpCAM [26], as most cancers of interest were epithelial in origin. However, EpCAM is lost during an EMT [64], leading to a misrepresentation of CTCs collected by this method. Indeed, even when captured with an EpCAM retrieval method, CTCs in breast and prostate cancer patients were found to co-express epithelial and mesenchymal markers in progressive disease [6], bringing up the question of how many EpCAM-negative mesenchymal cells were missed in the analysis. In response, other microfluidics-based methods of CTC capture have been adopted [65,66], although EpCAM-based methods still dominate patient diagnostics [67]. Non-specific capture methods have identified relatively equal populations of epithelial, intermediate, and mesenchymal CTCs, defined by EMT markers, such as E-cadherin and vimentin; however, multiple studies have found a correlation between high presence of mesenchymal CTCs and worsened patient prognosis [7,63]. Along with CTCs, microfluidics devices have also identified circulating tumor cell clusters (CTC clusters) which, although rare, are much more potent metastatic seeders than individual CTCs alone [14]. Under unbiased collection, identification and classification of CTCs and CTC clusters by microfluidics is a powerful diagnostic tool that can be combined with a multitude of other methods to understand EMP and its role in the metastatic cascade.Flow cytometry techniques can easily and readily detect cell populations expressing a combination of cell surface markers. Further, live cell populations can be sorted based on marker expression via fluorescence assisted cell sorting (FACS) for further study. This has made flow cytometry a very popular and easily applicable resource in many early studies defining E–M states. The invasive nature of cells undergoing an EMT elicited a natural link to cancer stem-cell like states, prompting these tumorigenic populations of cells to be initially isolated and described as CD44hi/CD24lo [22], ALDH+ [68] stem-like populations, and later linked to the EMT process [24,25]. Flow cytometry has provided a means of differentiating epithelial (EpCAM+/CD24hi/CD44-) and mesenchymal (EpCAM-/CD44hi/CD24-) in multiple cancer cell lines [69,70] as well as the breast [71] and prostate [53] mouse model to delineate differences between these states, understand the unique mechanisms that control EMT and MET, and determine their various roles in disease progression and the metastatic cascade. Further endeavors to increase the flow sorting sensitivity of E–M states has led to the discovery of novel EMT cell surface markers, such as CD104 (ITGB4) [23] as a supplement in addition to CD44/CD24 to define intermediate EMT states with cancer stem cells properties within human tumors, as well as combinations of EpCAM, CD51, CD61, and CD106 [28] to isolate multiple transitional intermediate/hybrid EMT states from mouse tumor models (Figure 1).These works and the application of flow cytometry established a link between the hybrid or intermediate state and increased stemness and decreased patient prognosis using several cell surface markers; however, this technique is unable to consider the expression of intracellular markers, such as vimentin or ZEB1, which require cell permeabilization. In addition, this technique can only be used in live tissue, making it more challenging to study patient samples, which are often archival formalin-fixed paraffin embedded (FFPE) tissue. Additionally, evidence from our recent study on EMT states suggests that canonical cell surface markers (CD44, CD104, and EpCAM) are not sufficient to separate distinct intermediate states from one another [59]. Indeed, Pastushenko et al. [28] relied upon the co-staining of four markers and precise gating strategies to adequately separate these states. Thus, flow cytometry presents an excellent approach for basic biological analysis of E–M plasticity but has few applications for direct use on archival patient tissues.Immunohistochemistry is the most common form of immunostaining and has been used for decades to detect and label antigens in tissue sections [72]. Hematoxylin and eosin (H&E) staining for DNA and proteins, respectively, is the principal method for histological assessment of tumor grade and histological subtype. In addition to assessing tumor grade and histologic subtype, IHC staining for other biomarkers is routinely performed by pathologists for certain tumors, such as hormone receptors, HER2, and Ki67 in breast cancer, to provide prognostic and predicative information, and to stratify tumors into molecular intrinsic subtypes [73,74].Immunofluorescence (IF) staining for EMT markers, particularly E-cadherin and Vimentin, has frequently been used alongside other methods for visualizing the co-expression of epithelial and mesenchymal markers as well as discerning the sub-cellular localization of proteins. However, it is rarely used as a comprehensive method for defining EMT states, owing, in part, to the limitation of fluorescence wavelengths that only allow visualization of a limited number of markers simultaneously. Efforts to combine changes in cell morphology with E-cadherin/Vimentin IF staining in a predictive EMT model have been partially successful in cell lines [75]; however, the application of this predictive model in vivo remains unclear. An immunofluorescence microscopy assay for cytoskeletal remodeling elements [76] has been successfully implemented as a readout for EMT disruption to screen a panel of transcription factor-targeting siRNAs to determine transcriptional nodes that control EMT [77], which can be useful for easy drug treatment and screening for future testable therapeutics targeting EMT. Moreover, recent work has combined fluorescent lineage tracing with intravital live microscopy to visualize and trace early and late EMT states in the primary tumor and metastatic sites, providing a much needed look inside the dynamic processes of tumor progression and EMT [13]. Combined immunofluorescence with other techniques, such as cell morphology or single-cell segmentation, has distinct advantages over bulk flow or sequencing by maintaining tumor architecture and spatial organization in the tissue. However, the limiting number of probes in classical IF presents the same drawbacks as flow cytometry and may be insufficient to describe the complexity of E–M states (Figure 1).RNA-based, and later chromatin-based, methods of assessing the EMT state have two main goals: to generate an EMT gene signature, or to characterize EMT gene networks across a spectrum of samples. This can be done in a variety of ways, although the goal is generally to further basic knowledge rather than apply directly to patients.Bulk RNA-seq has been repeatedly used to generate EMT gene signatures or “EMT scores” to help standardize and define entrance into an EMT [56] or partial EMT states in many cancers [57] and correlate that gene signature with poor patient prognosis. This is useful in understanding the connection between EMT gene signatures and patient prognosis as well as defining EMT states for new model systems [59]. However, many groups have put forth their own signature or method for ranking EMT [40,56,57,78], calling into question a standardized signature for ubiquitous use. In a more exploratory approach, RNA-seq has been used to interrogate EMP using either isolated clonal states within an EMT [28,59], or an EMT induction and withdrawal (MET) time course. These experiments served to delineate the specific gene networks that are active during the transition from epithelial to mesenchymal and back [77,79] and to help distinguish targetable transcriptional networks in aggressive or metastatic cell states. Combinations of EMT induction and siRNA knockdown of EMT target transcription factors identified control nodes, such as TEAD2, FOSL2, SP1, and others that had not been previously associated with EMT [77].Other approaches to visualize RNA expression, such as fluorescence in situ hybridization (FISH) probing for a panel of EMT markers, have had success in assessing EMP at a single-cell level in circulating tumor cells [15] before the more widely accepted single-cell sequencing approaches were robust enough to be used in this context.Single-cell RNA sequencing has been particularly vital in assessing EMT in tumors or cells where the bulk RNA signature may not be sufficient to describe the heterogeneous populations within each sample. This can be applied to CTCs [55], tumor cells [80], or as part of induced time course [79,81] to delineate EMP in these samples as well as interrogate EMP and heterogeneity at multiple points during metastasis, including the primary tumor, CTCs, and metastatic sites, to see how EMT states may work individually or cooperatively to promote metastasis.A leap in the field came from non-specific sequencing of accessible chromatin (ATAC-seq), which revealed large scale chromatin modification in response to progression through an EMT [28,82,83], indicating that large transcriptional shifts may be controlled through a combination of epigenetic and transcriptional regulation. Recently, multiple efforts have taken a multi-omics approach, combining RNA-seq and ATAC-seq to determine these combined epigenetic and transcriptional regulatory proteins, such as CTCF, the AP-1 complex, and the RUNX transcription factor family [59,84,85].While transcription-based approaches have provided a wealth of data and greatly contribute to the understanding of EMT, MET, and the metastatic cascade, the cost of sequencing, the processing times, and the inability to segregate tumor from stroma still make it inapplicable to assess patient samples for routine diagnosis. However, these comprehensive analyses have pinpointed specific EMT indicators for further and more directed approaches, such as multiplexed staining (Figure 1).Akin to immunohistochemistry, image-based methods of assessing tumors have distinct advantages for patient diagnosis. They are relatively easy, can be done with high throughput, and most importantly, retain spatial organization and heterogeneity of the original tumors. However, immunohistochemistry or even immunofluorescence struggles to describe the complexity of patient tumor states that may affect disease progression and metastasis, particularly epithelial to mesenchymal plasticity. Previous efforts have combined immunohisto-fluoresence for E-cadherin and Vimentin with high content screening (HCS), introducing a method that combines cell segmentation, morphological evaluation, and marker expression to determine nuanced EMT states within a tumor sample at a single-cell level [75]. This has been implemented with various other probes and image analysis software to combine immunofluorescence and morphological features into a reliable patient diagnostic tool [86,87]. However, this method is limited by the number of markers that can be used. Others, in efforts to combat this issue, have relied on mass cytometry time courses to map changes and co-expression of E-cadherin, Vimentin, CD44, CD24, and others in individual cells undergoing a TGF-β-induced EMT in lung cancer [58]. Similar methods have been used for multiplexed identification and stratification of heterogeneity in breast cancer patient samples with 35 different markers [88]. Even newer technologies, such as Nanostring DSP, present exciting new platforms for high-plex spatial imaging of RNA and/or proteins. This platform is becoming increasingly useful to deconvolute tumor heterogeneity and the tumor microenvironment (TME) of specific tumor types [89]. These approaches rely on precise image analysis software that has only recently become sensitive enough to reliably segment individual cells and deconvolute multiplexed staining approaches. Considering these technological advances, image-based approaches to quantify EMT progression in tumors are becoming more popular and easily implementable; these technologies have been reviewed extensively elsewhere [90]. Recently, we have employed a multiplexed, multi-round tyramide signal amplification (TSA) staining method using six canonical EMT markers that was used with cell segmentation and morphological features to define an EMT heterogeneity score and overall tumor EMT score in a model system of EMT, and further validated in a cohort of breast cancer patient samples [59]. Notably, this method was the first to reliably segment out stromal tissue, such as fibroblasts, which can surround tumors and frequently express Vimentin and ZEB1, mesenchymal markers that would skew an analysis of tumor composition. When implemented with widely practiced immunohistochemical approaches to determining tumor grade and composition, this method can help to elucidate the complexity of patient tumor heterogeneity and EMT state in the clinic to better inform prognosis and treatment regimes (Figure 1).Endeavors to characterize, quantify, and stratify epithelial–mesenchymal cell states in research models and patient samples has spanned decades. With each technological leap, the field gains more knowledge and insight into the markers and methods that can best and most simply stratify these phenotypic cell states. While no method is obsolete, some, such as immunohistochemistry, have made way for more complex and descriptive methods, such as multiplexed immunofluorescence. Ultimately, when patient diagnosis is the goal, approaches should be tailored for these specific needs, such as ease of use, number of samples at a time, cost, and accurate resolution of tumor tissue and individual cells. Flow cytometry and immunohistochemistry are simple and easy to implement, but lack the complexity and standardization required to reliably identify and score EMT states in patient samples from many different tumors. Transcriptional and chromatin-based methods provide this complexity and have pulled back the veil on the intricate transcriptional and chromatin regulatory networks that controlled epithelial–mesenchymal plasticity. However, they are difficult to implement on fresh samples and are quite cost prohibitive. For this reason, they remain a strong tool for use on test cohorts and in vitro or in vivo models of EMT but are unlikely to be adopted for routine diagnostics. Circulating tumor cells have been excellent diagnostic tools in patients for many years as they are easy to sample and provide a heterogeneous window into the tumor itself. CTCs also provide a background for testing many prognostic tools and have been tested out with many methods, past, present, and future. Image-based approaches have built upon what the field has learned about most descriptive and succinct E–M markers, as well as tissue and single-cell segmentation to create a robust tool to apply across many patient samples and in many contexts. In the future, such a method can be used to complement histopathological assessment in a clinical setting to provide a rapid and comprehensive analysis of E–M heterogeneity and the EMT tumor score to predict disease progression and inform treatment regimens.Conceptualization, M.S.B. and D.R.P.; writing—original draft preparation, M.S.B.; writing—review and editing, M.S.B., K.E.M. and D.R.P.; visualization, M.S.B.; supervision, D.R.P.; project administration, D.R.P.; funding acquisition, D.R.P. All authors have read and agreed to the published version of the manuscript. This research was supported by an Institutional Development Award (IDeA) from the National Institute of General Medical Sciences of the National Institutes of Health (under grant number P20GM104416), a Prouty Pilot Grant from Friends of the Norris Cotton Cancer Center, funding from The Elmer R. Pfefferkorn & Allan U. Munck Education and Research Fund at the Geisel School of Medicine at Dartmouth, an NCI Cancer Center Support Grant (5P30CA023108-40) and funding from the NIH 5R00CA201574-05 (to D.R.P.) The APC was funded by discretionary funds from the Norris Cotton Cancer Center.The authors declare no conflict of interest. Methods of Assessing EMT.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01139.txt

@@ -0,0 +1 @@
+Mouth cancer often results in poor outcomes and requires the use of state-of-the-art medical approaches to make its detection easy, individualized, and early. Liquid biopsy is a new and important medical approach to disease detection. This approach has been successfully used for mouth cancer detection and monitoring of treatment progress in many countries. Liquid biopsy is an attractive option for mouth cancer detection because it does not involve any invasive procedure and can be used on easily accessible body fluids, such as saliva and blood. Furthermore, there is evidence that this technology has some advantage over the normal tissue biopsy because it is not invasive, neither does it need any surgical expertise. Hence, we have focused on how easy or practical it would be, to employ the use of liquid biopsy on the African continent, as well as other low- and middle-income countries. We have discussed the different types of this technology in three main areas of focus, viz, what factors are important before, during and after collection of samples for liquid biopsy analysis, and what are the obstacles to routine use of this approach in resource-limited settings.An important driving force for precision and individualized medicine is the provision of tailor-made care for patients on an individual basis, in accordance with best evidence practice. Liquid biopsy(LB) has emerged as a critical tool for the early diagnosis of cancer and for treatment monitoring, but its clinical utility for oral squamous cell carcinoma (OSCC) requires more research and validation. Hence, in this review, we have discussed the current applications of LB and the practicality of its routine use in Africa; the potential advantages of LB over the conventional “gold-standard” of tissue biopsy; and finally, practical considerations were discussed in three parts: pre-analytic, analytic processing, and the statistical quality and postprocessing phases. Although it is imperative to establish clinically validated and standardized working guidelines for various aspects of LB sample collection, processing, and analysis for optimal and reliable use, manpower and technological infrastructures may also be an important factor to consider for the routine clinical application of LB for OSCC. LB is poised as a non-invasive precision tool for personalized oral cancer medicine, particularly for OSCC in Africa, when fully embraced. The promising application of different LB approaches using various downstream analyses such as released circulating tumor cells (CTCs), cell free DNA (cfDNA), microRNA (miRNA), messenger RNA (mRNA), and salivary exosomes were discussed. A better understanding of the diagnostic and therapeutic biomarkers of OSCC, using LB applications, would significantly reduce the cost, provide an opportunity for prompt detection and early treatment, and a method to adequately monitor the effectiveness of the therapy for OSCC, which typically presents with ominous prognosis.Reaching the goal of early and accurate diagnosis, as well as providing the best evidence-based treatment option for disease, is a key driving force for precision and individualized medicine [1]. A plethora of diagnostic methods have already been developed and, recently, liquid biopsy (LB) has shown increased global interest as a precision diagnostic tool in the field of cancer research. Cancer as a major global public health problem is emerging as a critical concern in Africa, where many cancer cases are diagnosed at late stages of the disease. This is due to factors such as limited knowledge and expertise of disease screening, limited diagnostic infrastructure, and for most patients, fear of surgery, poverty, lack of access to specialist care, and low educational level, among other factors—these are some of the key barriers to early presentation and cancer diagnosis among African populations [2]. Novel diagnostic tools such as LB could help in addressing these challenges due to its non-invasive nature, accuracy, and the fact that it does not require surgical facilities. LB has emerged as a rapid, reliable, and minimally invasive cancer screening solution, with high specificity and sensitivity for cancer diagnosis and monitoring. As demonstrated in developed countries, the high specificity and sensitivity of LB offers a promising diagnostic tool that would enhance screening capability and the potential for the early diagnosis of cancer cases in Africa; as well as probably reduce the incidence of morbidities and mortalities from cancer on the continent. In addition, the implementation of policies that allow easy access to valuable cancer diagnostic procedures, such as LB, can further help reduce the financial burden of late cancer management in many low-income countries in Africa. Oral squamous cell carcinoma (OSCC) ranks amongst the ten most prevalent cancers in the world with high morbidity and mortality rates [3]. This emphasizes the need for, and the importance of, screening programs and techniques for the early detection of malignancy. A lack of access to oral health care, which can lead to a delay in diagnosis has been reported to decrease the survival rates of OSCC in several low and middle-income countries including those in Africa [4]. The timely detection and diagnosis of OSCC may save lives by improving the survival rate, reducing treatment-related morbidities and improving the surveillance of recurrent cancer cases in these countries [5]. Therefore, it is vital to further understand how LB procedures are currently emerging and are used in low-and middle-income countries and, most importantly, to understand the blood biomarkers involved in oral cancer.For centuries, the use of tissue biopsies in cancer has enabled the histological characterization of the disease. Its application has provided insights into the genetic profile of tumors, allowing for good cancer management [6]. Notably, tissue biopsy remains the gold standard for diagnostic analyses in clinical settings. However, tissue biopsies involve invasive surgical techniques, cost, and tissue sample preparation. More importantly, tissue biopsies may not capture genetic heterogeneity within a tumor, and in intermetastatic tumor samples, thus affecting the accuracy of the test [7]. These challenges described make LB an appealing alternative, especially as a useful instrument in long-term management and prognostication. LBs can be used to investigate biological components in liquid forms in cancer patients for diagnosis, screening, and prognosis. LB may involve the analysis of released circulating tumor cells (CTCs) and circulating tumour DNA (ctDNA) in the blood or body fluid of a cancer patient [8]. These analytes are complementary biomarkers that present great potential for various cancer drug discovery platforms. Other analytes that can also be identified by using LB include circulating cell-free RNA (cfRNA), exosomes and platelets [9]. Liquid biopsy analytes can improve our understanding of tumor heterogeneity, and provide potentially better cancer diagnosis, treatment, and surveillance, as well as detecting drug resistance. Importantly, saliva, urine, pleural effusions, seminal plasma, sputum, cerebrospinal fluid, and stool samples are other physiological fluids that can be utilized for a LB in addition to blood [10,11].Fragmented, tumor-derived DNA that are not associated with cells (i.e., cell-free) and which are found within the circulatory system are known as ctDNA. They are the tumor derived part of circulating cell-free DNA (cfDNA), which refers to the total DNA shed into the blood and biological fluids during apoptosis and necrosis under physiologic and pathologic conditions [12]. Currently, ctDNA has been used in monitoring the therapeutic response and the detection of cancer relapse at early stages. Regarding treatment response and relapse detection, the identification of tumor-specific point mutations, promoter hypermethylations, and the identification of allelic imbalance using microsatellite markers analysis in ctDNA are helpful tools of assessment in patient management [13]. The ctDNA possesses short nucleic acid fragments of around 166 bp located in the plasma [14]. Healthy individuals have lower levels of cfDNA when compared with OSCC patients, which further increase as the cancer metastasizes, indicating its simultaneous use as a promising diagnostic and prognostic biomarker [15]. However, increased cfDNA is not specific for cancer, and in individual patients, there is no specific cut-off value which can be attributed to the tumor in quantification as ctDNA. This limitation can be overcome by evaluating tumor specific alterations such as methylations and mutations [16]. The potential for using ctDNA in all stages of head and neck squamous cell carcinoma (HNSCC) diagnosis and management was highlighted in a recent review—this included screening and early detection; prognosis and the detection of minimal residual disease; the characterization of mutational landscape; precision medicine and treatment selection; and treatment monitoring and identification of resistant clones, based on experience with HNSCC of other sites using ctDNA analysis [17]. Several studies have been developing techniques to detect cfDNA (or ctDNA), but there is still a lack of a standardized method, which is essential for its clinical application together with the need to reduce the cost of analysis. Studies have identified ctDNA as a diagnostic biomarker and this has been used in many diseases [18]. Although these studies demonstrated that circulating cfDNA can screen for cancer; when testing for ctDNA, the procedure requires additional investigation and an acceptable cost before these can be applicable for use as a diagnostic or therapeutic tool, especially in African and other resource-limited settings.CTCs are released into circulation by the primary tumor (i.e., they are of tumor origin) as a result of spontaneous or iatrogenic factors. They are believed to share a similar profile with the somatic mutations and genomic rearrangements present in the primary tumor [19]. They are able to reflect the tumor heterogeneity, which may be missed in surgical biopsies. This makes them good candidates for understanding tumor mutational profiles without allowing patients to go through invasive tissue biopsy. A major problem with CTC’s use in patient management is their extremely low number in peripheral circulation (1–100 CTCs per 1 billion peripheral blood cells). They need to be enriched and separated for detection and analysis by often tedious and expensive methods [20,21]. Increasing levels of CTCs have been reported to be associated with poor prognosis and distant metastasis in several forms of cancer. Survival rate in OSCC, and other cancers such as breast, lung, prostate and ovarian cancer, has been linked with levels of CTCs, indicating their benefit in the early screening and treatment monitoring of cancer [22]. Partridge et al. [23] evaluated the levels of disseminated tumor cells preoperatively and intraoperatively in both blood and bone marrow from 40 patients with OSCC. They found a high risk of loco-regional recurrence and distant metastasis associated with the presence of CTCs. In 2015, Oliveira-Costa et al., provided more knowledge regarding CTC biology for OSCC by analyzing the gene expression profile of OSCC tumors to identify biomarkers that decreased or increased during tumor progression [24]. Their results showed that programmed death-ligand 1 (PD-L1), HOXB9 and ZNF813 expression in OSCC-derived CTCs was increased, while B Cell Linker (BLNK) expression decreased. In summary, the investigators reported that PD-L1 is a prognostic factor in OSCC as expressed in patients CTCs and provides insights for the development of an anti-PD-L1 therapy for OSCC patients. Due to PD-L1 inducing an exhaustion state in T-cells and reducing the capability of a T-cell-mediated response, they hypothesized that OSCC patients could benefit from anti-PD-L1 therapy. Therefore, the results reported in this study emphasized the role of CTCs as an independent prognostic marker in OSCC. Recurrent assessments of CTC levels have also been used in studies to demonstrate the potential utility of CTCs in disease monitoring before, during, and after therapy, Inhestern et al. [25] examined and evaluated CTC counts in blood samples from 40 patients with OSCC. Apart from its potential as a prognostic biomarker, there is interest in investigating the role of CTCs in regulating disease behavior. The checkpoint inhibitors that block the PD-1/PD-L1 immune checkpoint pathway on CTCs and stimulate the immune system to remove CTCs in circulation may reduce the risk of metastasis and disease recurrence. In patients with OSCC, PD-L1 overexpression in CTCs was identified and utilized to monitor the treatment response [24]. Numerous studies [26,27,28,29,30] showed that LB procedures, such as CTCs, can be utilized as a cutting-edge technology that may improve the detection and monitoring of cancer using a small amount of blood samples. CTCs assessment using LB as a cancer biomarker promises to be low cost for the management of patients with OSCC. Exosomes are bioactive vesicles with diameters ranging between 40–150 nm that are used for analyzing LBs. A miRNA expression profile has shown that circulating exosomal miR-21 was associated with hypoxic tumor and metastasis in the lymph node in people with OSCC [31]. The detection of miRNA biomarkers in both the plasma and tumors of patients with squamous cell carcinoma of the tongue highlights the significance of free and exosomal miRNAs as potential diagnostic biomarkers for tongue cancer. In addition, packaged circulating miRNAs in protein complexes or encapsulated within microvesicles are protected against the activity of blood RNAses, and represent a more dependable approach for the assessment of circulating tumor-miRNA signatures [10]. More so, exosomes in the tumor microenvironment have been implicated in increasing levels of the transforming growth factor-B (TGF-β) pathway; thus, increasing drug resistance and tumor growth in OSCC patients. Exosomal chemokine-like factor (CKLF)-like MARVEL transmembrane domain-containing 6 (CMTM6) of OSCC cells aid the polarization of alternatively activated macrophages (M2) via activation of the signaling of ERK1/2 in macrophages [32]. Indeed, the classically activated macrophages possess anti-tumor properties; conversely, the M2 functions as a pro-tumor, aiding in the development and progression of the tumor [33]. Exosomes are released by different types of cells into numerous biological fluids such as amniotic fluid, cerebrospinal fluid, lymph, bile, ascites, tears, breast milk, urine, semen, blood, and saliva, both in healthy and diseased conditions [34,35,36,37]. Two research groups have demonstrated that exosomes are present in the tumor microenvironment, demonstrating its importance in tumorigenesis, tumor invasion, and metastasis, since they can possess an anti-tumor function or promote tumor progression [38,39]. In OSCC, exosomes have been shown to be key components in the tumour microenvironment, increasing the TGF signaling pathway, which contributes to the progression and drug resistance of OSCC [40]. Limited available evidence may suggest a potential discriminatory biomarker role of exosomes, between active OSCC disease patients and cured OSCC patients. In addition, a study by Zlotogorski-Hurvitz et al. [41] morphologically characterized oral fluid-derived exosomes in OSCC. The potential scope of the diagnostic and prognostic application of exosomes in oral cancer has been described elsewhere [32]. Additionally, the role of some exosomal miRNA (e.g., miR-223, miR-101-3p, miR-338 and miR-34a-5p) as tumor suppressors and the robust potential of exosomes for therapeutic drug delivery to the tumor for effective treatment or to improve prognosis has been highlighted in another recent review [34].LBs find and evaluate several biomarkers, such as mRNA biomarkers, pro-inflammatory cytokines, and metabolites in the saliva, urine, and plasma of OSCC patients. mRNA biomarkers from saliva for use in the early diagnosis of oral cancers were recently described by Oh and colleagues [35]. Thirty candidate genes related to cancer previously reported in the literature were selected. Thirty-three OSCC patients and 34 non-tumor controls had their saliva samples taken and the mRNA levels of six genes CYP27A1, NAB2, collagen type III alpha 1 (COL3A1), monoamine oxidase B (MAOB), nuclear pore complex interacting protein B4 (NPIPB4), and sialic acid acetyltransferase (SIAE) were considerably lower in the saliva of OSCC patients. The combination of SIAE and CYP27A1 had an AUC of 0.84, which was considered good. In the under 60 group, the AUC of MAOB–NAB2 was more prognostic of OSCC (AUC, 0.91; specificity, 0.86; and sensitivity, 0.92) than any other transcript combination. The results from this study suggested salivary mRNAs were useful biomarkers for early OSCC diagnosis, especially in individuals under 60 years old. Lu et al. [36] found substantially increased expression of plasma miR-196a and miR-196b in patients with cancerous and precancerous oral cavity lesions, with excellent sensitivity and specificity compared to the normal controls. Furthermore, Liu et al. [37] found a substantial increase in plasma miR-31 in OSCC patients, and a significant decrease following tumor excision, indicating that miRNAs might be used for diagnostic and treatment monitoring purposes.Saliva allows for a non-invasive LB; it is easily accessible and has a large number of biomarkers for illnesses, as well as pre-symptomatic and health status indicators. The saliva biofluid contains RNA and DNA molecules, cytokines, extracellular vesicles (EVs), and circulating and tissue-derived cells, which are novel biomarkers or indicators [38,39,40]. Salivary biomarkers may be effective for the early detection of OSCC since they are physically accessible to the mouth cavity [38]. In saliva and tissue samples from individuals with OSCC and oral potentially malignant diseases, NF-κB-dependent cytokines, and pro-inflammatory cytokines (IL-6, IL-1α, IL-8, and TNF-α) were assessed [41]. NF-κB-dependent cytokines, matrix metalloproteinases (MMPs) and pro-inflammatory cytokines in saliva may be involved in the relationship between oral malignancies and aging, involving the senescence-associated secretory phenotype (SASP), and inflammatory diseases, such as oral mucosal ulcers and periodontitis [42]. Metabolome biomarkers in OSCC were also investigated using nuclear magnetic resonance (NMR) metabolomics to detect high-risk patients with an extranodal extension [43]. Recent research has revealed that EVs, including exosomes (small vesicles) and bigger vesicles, are released into the saliva from oral cancer lesions and/or oral tissues [44]. Meanwhile, many biomarker proteins of oral malignancies, such as heat shock proteins (HSPs) [45,46,47], and epidermal growth factor receptor (EGFR) [48], have been discovered in EVs. Thus, microRNAs in EVs of saliva could also be important molecules for early detection of oral malignancies distinct from aging [49], periodontitis [49,50], and state of health [51].More sensitive emerging technologies are now employed for LB analysis. These technologies include beads, emulsion, amplification, and magnetics (BEAMing), digital droplet PCR (ddPCR) and next generation sequencing (NGS) [36]. In respect of experimental and clinical applications, BEAMing, ddPCR and NGS have separate applications, and they sometimes complement each other for the examination of a tumor at molecular level [36]. ddPCR is very reliable and accurate for the investigation of genetic alteration in various cancers due to its very sensitive nature [37]. However, this method is challenged by a low multiplexing capacity, but efforts are in the pipeline towards designing multiplexed strategies to reduce experimental artefacts [37]. LBs can be used to monitor OSCC treatment responses and tumor evolution in real-time and improve the prognostic and diagnostic potential for OSCC [10,38]. One of the essential benefits of analyzing LBs in OSCC is that they give a tailored snapshot of primary and metastatic tumors at different periods, allowing clinicians to detect early signs of disease recurrence or resistance to therapy and assisting them in their therapeutic decisions [10]. As a result, by employing LBs, a molecular profile for each patient can be acquired. Different tumor subtypes may be useful in supplementing the tumor, node, and metastasis (TNM) staging approach [10]. A summary of the applications of LB in OSCC can be found in Table 1.Despite the progress made by oral LB in different parts of the world, its practical application in African and other resource-limited settings remains a major challenge. Basically, tumor components, including ctDNA, ctRNA, CTCs and/or extracellular vesicles (exosomes), shed into the circulation from the tumor site, are increasingly becoming clinically useful in the detection and monitoring of tumor or cancer progression [10]. These biopsies are employed for early detection of the circulating tumor origins from bodily fluid. However, the process of employing these biopsies has been faced with several challenges, which have hampered their optimal treatment applications. For instance, although oral swabbing can be used for tumor surveillance after treatment, it is very difficult to apply this screen to a high-risk population without tumor relevant lesions.Moreover, oral sample may not be useful in differentiating the origin of the tumor cells in the circulation [52]. In addition, cfDNA comprises mutated and non-mutated tumors and normal DNA; thus, depending on the types and stages of the tumor, differentiating ctDNA from the mix could be burdensome. In some instances, a very small amount (≤0.01%) of ctDNA is used for variant genotyping. This often occurs where there is an extremely low frequency of CTCs, making it difficult to detect ctDNA at the early stage of cancer, and further complicating the detection and recovery processes [30]. The inability to differentiate and specifically target tumor components or associated biomarkers from already existing non-tumor cells in circulation is a critical issue facing all LBs, which might have delayed their optimal clinical use [53]. For example, trying to use ctDNA to capture the full heterogeneity of cancer cells may also introduce the possibility of capturing DNA of normal cells [54], which may consequently increase the chances of error and false-positive results. Due to the heterogeneity of tumor cells, there is a likelihood of genetic alteration of the CTCs detected in bio-fluid circulation compared to the original tumor profile, decreasing the accuracy of the finding [55]. LB may never reach its full potential because detecting CTCs (which is the major biomarker targeted during a LB) requires a huge amount of blood or complex enrichment media to achieve an optimal result and an acceptable level of specificity and sensitivity [56]. Regarding the LBs, specificity and sensitivity cannot be overemphasized as they are critical to validating the result’s accuracy, efficacy, and reliability. In the assessment of CTCs in LB, the use of epithelial cells as sole biomarkers for the enrichment of this assay poses a serious challenge, especially when epithelial cells have lost their physical characteristics and specificity after undergoing epithelial-to-mesenchymal-transition (EMT) and gained mesenchymal properties [57]. This makes it almost impossible for them to be detected by the typical methods of CTC isolation; hence, compounding the issues of sensitivity and specificity of LB. Again, the reported high level of epithelial cells in the circulation of patients suffering from benign diseases is an indication that the use of the epithelial cell adhesion molecule (EpCam) to capture CTC may introduce unwanted errors and misdiagnosis [58]. Sample collection, storage, shipping, and transportation are critical for the stability of the collected specimen. If they are not well handled, it could lead to the degradation of the sample, hence producing varying results. The Ethylenediaminetetraacetic acid (EDTA) used to maintain the longevity and stability of blood LBs has a high incidence of genomic DNA contamination when stored for a long time [59]. Due to poor transport facilities in the major parts of African countries, it is unlikely that the reagent would reach the rural communities in good condition, which would hamper its clinical application efficacy. The reproducibility issue of biomarkers is due to the lack of standardization guidelines that give rise to the variability of results as reported in vesicle counting of extracellular vesicles, which reduces the reliability and reproducibility. Therefore, the results may not be reproducible or comparable in various study cohorts [60].In an African context, many obstacles could prove to be wedging blocks that prevent the practical implementation of LB in OSCC in this continent. Thus, so far, only three African countries have adapted and authorized or commercialized the use of these techniques in the public domain. These countries are South Africa, Tunisia, and Kenya, and this could be due to weighing the financial burden of cost, approximately US$7000 per test, which can only be afforded by high-income earners who form less than 10% of the population of most African countries [61]. It will become more challenging to implement in low- and middle-income countries in sub-Saharan Africa with a Gross Domestic Product (GDP) of less than US$15 billion. Moreover, considering that most of the available LBs have not been tested in African populations, it may be harder for physicians to ascertain their efficacy in Africa due to environmental and genetic variations. In addition to the inadequate infrastructural facilities, technical know-how, limited skilled personnel, poor government policy, a lack of many adequate research institutes, and religious and cultural practices have all slowed scientific innovation and progress in Africa [62]. Further, the coronavirus pandemic has negatively impacted the economy of many African countries and this could further delay the practical implementation of oral LB in Africa. Furthermore, despite its evidence-based theranostic potential to improve cancer management, LB is prone to several analytical discrepancies, which until resolved, could limit its applicability in clinical practice. Inconsistencies and errors may be notable at any of the stages in the workflow, leading to sample misevaluation. These methodological inconsistencies may pose a risk to patients due to a discrepancy in cancer management and the unnecessarily high costs to patients, laboratories, and hospitals [63]. Therefore, consistency in a clinical setting is critical for optimal patient safety, the accuracy and precision of laboratory tests and, overall, the routine applicability of LB in clinical settings.In the pre-analytical phase, a substantial part of the sample workflow occurs outside a controlled laboratory setting [64,65]. Previous reports have highlighted that this phase is subject to the most errors in the sample evaluation workflow [63], and could limit the integrity of the sample and the data quality [66,67]. Study-specific variations in ctDNA assays and insufficient ctDNA recovery limit the validation of plasma ctDNA based LBs for cancer screening under a clinical setup [68]. Multiple studies have demonstrated that the quality of cfDNA and ctDNA could be impacted by the type of sample collection tube used, optimal storage conditions, and the time-lapse between sample collection and sample processing [65,69,70,71]. As a result, the optimization of sample collection and processing is dependent on the type of collection tube used and is critical for optimal downstream data output. Serum isolating standard EDTA tubes have been linked with leukocyte lysis that could increase genomic DNA concentration and potentially reduce the ease of access to ctDNA, which is often available in small quantities [64,65,72]. To minimize the risk of genomic DNA contamination and dilution of ctDNA and cfDNA, optimizing the time-lapse between sample collection and the initiation of sample processing, as well as the sample storage temperatures during transportation is essential [71]. Genomic DNA contamination of cfDNA could be minimized by sampling whole blood [65], and potentially saliva, in cfDNA stabilizing collection tubes containing fixatives. These tubes could improve the recovery of high-quality cfDNA with the added advantage of a prolonged storage time and variable shipping temperature [71,72,73]. However, these specialized tubes could be more costly than the EDTA tubes and, therefore, may not be easily implementable in resource-constrained settings such as in developing African nations. Similar to the pre-analytical stage, sample processing is a consolidated effort of many laboratory aspects aiming to produce high-quality clinical data. Easily identifiable is a varying range of methodologies including reagents and applied technologies (including different instruments) used by separate laboratories to achieve the same goal of generating high-quality clinical data [74]. Different laboratories may use different ‘clinically validated’ consumables yet with differing clinical results, particularly when different commercial suppliers are used for supposedly similar clinical assessments [75]. This poses the risk of differing and possibly incorrect interpretations of clinical data, and thus puts patients at risk of incorrect cancer management plans [75].Several LB analytical methods with varying sensitivities and specificities are prone to laboratory-specific variations [65]. In the processing phase, detection technologies used to process LB samples are prone to variations that could raise inconsistencies. In addition, most of these technologies are expensive and cannot be afforded by most low- and middle-income countries in Africa.Taking NGS assays as examples, despite their notable success in advancing the application of LBs in clinical settings, technical heterogeneity of these assays within the same studies or amongst different studies have been described as potential limitations to the implementation of NGS in routine clinical practice [76]. In addition, there are no consensus-based standard molecular coverage parameters affecting the sensitivity for NGS. Under these conditions, data output interpretation is often site-specific, with a heightened risk of false-negative results [76,77,78]. Similarly, ddPCR has been praised for its analytical strengths over real-time quantitative PCR [79] in addition to its high sensitivity, which is comparable to deep sequencing [80,81]. However, except for intralaboratory validations, no consensus-based analytical and clinical validation guidelines for cfDNA assays on ddPCR exist [79]. Consequently, the implementation of this technology is often site or laboratory-specific, and its output could be potentially variable across several laboratories. Inconsistencies in ddPCR could, therefore, influence the interpretation of the output data, resulting in inconsistent patient management depending on the site of the clinical sample evaluation. As such, these inconsistencies require careful control through expert agreement, standardized validation, and optimization practices in clinical settings [65,77,82].The use of these technologies need to be refined and tailored to address the operational challenges, which currently limit their application in Africa [83]. The financial burden of using these molecular technologies deserves some consideration. The cost concerning technical and staff equipment, including a variety of professionals such as molecular and computational biologists, genetic counselors, and specialized clinicians, are considerably high and can be unaffordable for many low-income countries on the continent [84].The optimization of molecular data analyses would improve reproducibility while maintaining high-quality clinical inferences. Laboratory protocol and analytical workflow standardization remain the cornerstone of clinical laboratory practice to ensure consistently reproducible and accurate clinical data output for optimal patient management in OSCC cancer care [85]. The reproducibility and reliability of laboratory results can be improved by the standardization of sample processing workflow, through expert consensus. These can be achieved via large-scale clinical trials, which validates the critical steps required to improve the reliability, accuracy, and consistency of clinical data at acceptable methodology specificity and sensitivity levels [82,86]. Moreover, supporting the need for further research validation, LB in oral cancer is an emerging field that is still limited by the lack of knowledge on sensitive and specific circulating biomarkers for oral cancer detection [10]. Altogether, it is imperative to establish clinically validated working guidelines for various aspects of LB sample collection, processing, and analysis for optimal and reliable use of LBs and their associated technologies in routine clinical care (see Figure 1).LB has emerged as a potential non-invasive approach in precision, personalized cancer medicine, particularly for OSCC. Different studies have demonstrated the benefit of OSCC LB, using various downstream analysis such as CTCs, cfDNA, miRNA, mRNA, Long non-coding RNA (lncRNA) and salivary exosomes, and have shown promise for early OSCC diagnosis, treatment monitoring, and detection of occult disease and relapse [2]. However, many of these downstream LB analyses are yet to be routinely applied in clinics for oral cancer. In Africa, and especially in sub-Saharan countries, the impact of LBs in the clinical setting is still very limited, requiring further research for effective implementation. Further discovery of a robust panel of sensitive and specific circulating biomarkers for OSCC at different stages would help, not only policymakers in decision-making, but also clinicians to improve diagnosis and prognosis. A better understanding of OSCC biomarkers would be key for the development of effective therapies for the management of oral cancer. The continent needs to quickly improve the management of oral cancer, and since LB for OSCC is in its infancy, researchers working in oral cancer should put their efforts into performing large, but also prospective multicenter studies that investigate the role of CTCs, ctDNA, cfDNA, miRNA, incRNAs and exosomes in oral cancer. Better use of LBs in oral cancer can help in the monitoring and surveillance for post-treatment recurrence as well as the detection of new cancer cases. Thus, due to the cross-cutting low socio-economic status of populations, Africa needs a reproducible and non-invasive LB that could provide the basis for individualized diagnosis, therapeutic strategies, and precision management of OSCC. Although the advantages of LBs in the evaluation of OSCC prognosis, risk stratification, or its use as a tool for precision medicine development, has been discussed here, a very challenging aspect is its reproducibility across multiple centers, to establish a standard guideline for LB-based cancer diagnostics. The development of LB databases such as liqDB [87], ctcRbase [88,89], and BloodPAC Data Commons (BPDC) [90], has helped to overcome such technical issues and has encouraged data science research to address the practical and conceptual reproducibility problems associated with the use of LB [91]. Hence, the development of an OSCC LB database in Africa and other resource-limited settings, could potentially ameliorate the diagnostic and treatment burden of OSCC in these regions.Conceptualization, H.A.A. and I.O.B.; methodology, H.A.A., R.T.A., N.M.F., A.O.A. and I.O.B.; data curation, H.A.A., A.O.A. and I.O.B.; resources, H.A.A., A.O.A. and I.O.B.; writing—original draft preparation, H.A.A., R.T.A., N.M.F., T.A.A., P.C.I. and F.M.; writing—review and editing, H.A.A., R.T.A., N.M.F., T.A.A., P.C.I., I.O.B., A.O.A. and F.M., visualization, H.A.A., R.T.A., N.M.F., T.A.A., P.C.I., I.O.B., A.O.A. and F.M.; funding acquisition, H.A.A., I.O.B., A.O.A. All authors have read and agreed to the published version of the manuscript.This research received no external funding.Not applicable.Not applicable.The data presented in this study are available in article.H.A.A. thanks the South African Medical Research Council (SAMRC) for a mid-career scientist and self-initiated research grant; and the South African National Research Foundation (NRF) for research development grants for rated researchers. I.O.B. acknowledges the support of Prince Naif Health Research Center (PNHRC), Saudi Arabia.The authors declare that they have no conflicts of interest.Challenges associated with the implementation of liquid biopsy in poorly resourced settings. Created in BioRender.com, accessed on 21 July 2021).Some evidence of the applications of liquid biopsy in oral cancer.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01140.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Typical benign nevi and advanced melanomas can be easily discriminated, but there are still some melanocytic lesions where even experts are not sure about the correct diagnosis and degree of malignity. The high penetration depth of optical coherence tomography (OCT) allows an assessment of tumor thickness of the lesion precisely, but without cellular resolution the differentiation of melanocytic lesions remains difficult. On the other hand, reflectance confocal microscopy (RCM) allows for very good morphological identification of either a nevus or a melanoma, but cannot show the infiltration depth of the lesion because of its low penetration depth. Since the new device of line-field confocal optical coherence tomography (LC-OCT) technically closes the gap between these other two devices, in this study, we wanted to examine if it is possible to differentiate between nevi and melanomas with LC-OCT, and which criteria are the most important for it.Until now, the clinical differentiation between a nevus and a melanoma is still challenging in some cases. Line-field confocal optical coherence tomography (LC-OCT) is a new tool with the aim to change that. The aim of the study was to evaluate LC-OCT for the discrimination between nevi and melanomas. A total of 84 melanocytic lesions were examined with LC-OCT and 36 were also imaged with RCM. The observers recorded the diagnoses, and the presence or absence of the 18 most common imaging parameters for melanocytic lesions, nevi, and melanomas in the LC-OCT images. Their confidence in diagnosis and the image quality of LC-OCT and RCM were evaluated. The most useful criteria, the sensitivity and specificity of LC-OCT vs. RCM vs. histology, to differentiate a (dysplastic) nevus from a melanoma were analyzed. Good image quality correlated with better diagnostic performance (Spearman correlation: 0.4). LC-OCT had a 93% sensitivity and 100% specificity compared to RCM (93% sensitivity, 95% specificity) for diagnosing a melanoma (vs. all types of nevi). No difference in performance between RCM and LC-OCT was observed (McNemar’s p value = 1). Both devices falsely diagnosed dysplastic nevi as non-dysplastic (43% sensitivity for dysplastic nevus diagnosis). The most significant criteria for diagnosing a melanoma with LC-OCT were irregular honeycombed patterns (92% occurrence rate; 31.7 odds ratio (OR)), the presence of pagetoid spread (89% occurrence rate; 23.6 OR) and the absence of dermal nests (23% occurrence rate, 0.02 OR). In conclusion LC-OCT is useful for the discrimination between melanomas and nevi.Optical coherence tomography (OCT) and reflectance confocal microscopy (RCM) already play important roles in routine non-invasive skin cancer diagnosis [1]. OCT is well established for non-melanoma skin cancer (NMSC), whereas RCM is most commonly used for pigmented lesions [2,3,4]. With OCT it is possible to create vertical images until mid to deep dermal level, to reconstruct 3D images, measure tumor thickness, monitor topically treated lesions over the course of time, and even to distinguish BCC subtypes, but lacks cellular resolution [2,3]. This is why melanocytic lesions cannot be visualized very well with OCT. With the high resolution of RCM, pigmented tumors can be examined horizontally in cellular detail. Since the penetration depth of RCM is limited, it is only possible to measure from the stratum corneum to the stratum papillare of the dermis [4,5]. The new line-field confocal optical coherence tomography (LC-OCT) method offers the combination of the advantages of both devices—a high penetration depth like OCT and a similar high resolution like RCM, together with the visualization of images in horizontal and vertical views [6,7,8,9,10,11].Therefore, LC-OCT was used in this study to investigate melanocytic lesions and to find out the most common and most useful parameters for diagnosis, especially for the differentiation between (dysplastic) nevi and melanomas. The improved distinction between nevus and melanoma with LC-OCT allows a fast, in vivo, and non-invasive diagnosis. If the lesion can be categorized as a nevus immediately with LC-OCT, no excision is necessary. In addition to the benefit to the patient that no invasive procedure is needed, this new technique will help to avoid unnecessary surgeries, leading to lower healthy insurance costs, as well as to offering more free capacities for other necessary melanoma surgeries. This study assesses the influence of image quality on diagnostic performance with LC-OCT and RCM and determines the potential of LC-OCT (vs. RCM vs. histology) to identify a melanoma vs. a nevus reliably and accurately. To our knowledge, this is the first work about melanocytic lesions, especially concerning the differentiation of nevi and melanomas with the new LC-OCT device.The LC-OCT device used in our study is based on a time-domain-OCT (TD-OCT), which takes several A-scans from the skin’s surface down to a maximal depth of 500 µm for the acquisition of B-scans, while constantly refocusing. It has an axial resolution of 1.1 µm and a lateral resolution of 1.3 µm. The LC-OCT technique consists of a two-beam interference microscope, with a laser source of 800 nm wavelength with a continuous spectrum and a line camera as a photodetector. The laser classification is 1 M according to EN 60825-1. Like in conventional OCT, the images are depicted in grey scale. The CE-marked LC-OCT deepLiveTM (DAMAE Medical, Paris, France) is a mobile central unit with a monitor and a handheld probe. To generate high quality images and reduce the optical index between the glass plate of the probe and skin layers, paraffin oil was used. The device is non-invasive, painless, and is applied without pressure. LC-OCT has three available modes to create images in real time: vertical (en-coupe) sections, as in OCT and histology, horizontal (en-face) images as in RCM, and 3D images. The field of view for both the vertical and horizontal sections is 1.2 mm × 0.5 mm. The 3D images can be taken as horizontal stacks from the top of the skin with steps of 1 µm. Short videos can also be acquired with up to 26, 16, or 8 frames/s (basic, high definition, ultra-high definition). Moreover, a dermoscopic image (resolution 5 µm, field of view 2.5 mm) is taken simultaneously, which makes it possible to navigate in the lesion and choose the correct position. Captured images can be exported and saved globally or as a single patient/lesion file in TIFF, DICOM, or JPEG formats. The additional program 3DSlicer (The Slicer Community, Open-Source Software) allows the reconstruction of 3D cubes of the horizontal LC-OCT stacks [9,10].In our study we used commercially distributed RCM devices (VivaScope® 1500 and VivaScope® 1500/3000 Combo, Mavig GmbH, Munich, Germany), which have a penetration depth of 300 µm. Both have a lateral resolution of 1 µm and an axial resolution between 3–5 µm. The light source is an 830 nm diode laser. The 1500 device has an integrated dermoscopic camera for navigation, creates image sizes of 500 µm × 500 µm, and builds mosaics of single greyscale pictures up to 8 mm × 8 mm (VivaBlock®). The other mode, which we used in the study for the comparison with LC-OCT, is the VivaStack®, where single images can be taken from the skin’s surface to the stratum papillare in several steps. The conventional OCT used in this study is the commercially available, handheld-based OCT device VivoSight® (Michelson Diagnostics Ltd., Maidstone, Kent, UK). We additionally measured 30 of all the melanocytic lesions with conventional OCT for comparative reasons. The OCT acquires images of 6 mm × 6 mm with 1.5 mm detection depth. It has a lateral resolution of 7.5 µm and an axial resolution of 10 µm, and contains a laser source of 1305 nm. With the addition of dynamic OCT, blood flow and blood vessels can be visualized. Further details on the devices are described elsewhere [2,12].Patients with suspicious melanocytic lesions or nevi that were planned for excision were enrolled in the study at the Department of Dermatology and Allergology at the University Hospital of Augsburg and at the University Hospital of the Ludwig Maximilian University, Munich in Germany between November 2019 and January 2021. The study was approved by the local ethics committee (No. 17-699) and conducted according to the principles of the Declaration of Helsinki and international guidelines concerning human studies. Written informed consent was obtained from all patients prior to inclusion into the study.A clinical examination was done to identify the lesion as suspicious for a melanoma or a (dysplastic) nevus prior to excision. Clinical and dermoscopic images of each lesion were acquired using Fotofinder® (FotoFinder Systems GmbH, Bad Birnbach, Germany), DermoGenius 2® (DermoScan GmbH, Regensburg, Germany), ILLUCO IDS-1100 (DermoScan GmbH, Regensburg, Germany), or an iPhone 12 Pro Camera (Apple Inc., Cupertino, CA, USA). After dermoscopy, the lesions were scanned with LC-OCT (deepLiveTM, DAMAE Medical, Paris, France) in horizontal, vertical, and 3D. After that, OCT images (vertical and dynamic en-face) as well as RCM pictures were taken (VivaStacks® and VivaBlocks®). Horizontal LC-OCT and RCM images were taken at the level of the stratum corneum, the epidermis, the dermo-epidermal junction (DEJ), and the papillary dermis. For measurements with RCM and LC-OCT, a few drops of paraffin oil were used before the scan. OCT examinations did not require preparation of the skin. After the examination, a biopsy, shave, or excision of the lesion was taken and sent for histological analysis. All lesions were compared with standard histology in haematoxylin/eosin. After all images had been reviewed, the observers graded LC-OCT and RCM image quality during a consensus meeting with semi-quantitative scores as poor (3), acceptable (2), good (1), or excellent (0), and the confidence level from low (3), medium (2), high (1), to very high (0). After histology, the final diagnosis was noted. Both centers are regular users of OCT, RCM, and LC-OCT. Both centers had at least 3 months of practical experience and training with LC-OCT before the study. In addition to the well-known clinical and dermoscopic diagnostic criteria for melanocytic lesions, we used the following patterns for the different methods: LC-OCT criteria in horizontal and vertical view (see Tables S1 and S2) and RCM parameters (see Tables S3 and S4). In 30 cases, conventional OCT images were also recorded for demonstration purposes. Since RCM is superior to OCT in diagnosing melanocytic lesions and OCT cannot show the cellular details of melanomas and nevi, OCT criteria were not evaluated and OCT was not included in the systematic study [13,14].For the collection of data, Microsoft® Office Excel® for Mac 2021 was used. Statistical analysis was performed with R version 4.0.0 (R Foundation for Statistical Computing; Vienna, Austria). The accuracy, specificity, sensitivity of each technique, PPV (percentage of positive diagnoses that were correct), and NPV (percentage of negative diagnoses that were correct) were calculated. The specificity and sensitivity of LC-OCT for the diagnosis of a melanoma were compared with the specificity and sensitivity of RCM, and then both devices with histology using McNemar’s test, which considers the paired nature of the data. Furthermore, the investigators’ responses concerning the recognized parameters for melanocytic lesions, the diagnoses, the image quality, and confidence level were noted. For the intra-method correlation between image quality and confidence level, Spearman’s correlation coefficients (r) were calculated. A p-value < 0.05 was considered as statistically significant. When the data failed the normality test, a Wilcoxon test for paired tests was used to compare image quality and confidence level between LC-OCT and RCM. We also performed univariate and multivariate logistic regression analyses for the criteria that were more useful in discriminating a melanoma from a nevus (dysplastic or not), and used a backward elimination approach. A total of 75 patients with 84 melanocytic lesions were evaluated with LC-OCT before excision. A total of 36 of the 84 lesions were also measured with RCM. Five patients had two lesions, one patient had five lesions, and the others had one pigmented lesion. The mean age of the patients was 51 years. Lesions were mainly located on the trunk (57.1%), limb (25%), and the head (17.9%). Histology was available for all 84 lesions. Histology identified 42 (50%) of 84 lesions as banal nevi, 13 (15.5%) as dysplastic nevi and 28 (33.3%) as melanomas. A total of 65.5% of the lesions were histologically diagnosed as nevi (dysplastic or not). Furthermore, 82.1% of the lesions were completely excised, 16.7% were biopsied, and 1.2% were shaved. One lesion was histologically diagnosed as a pigmented actinic keratosis and was therefore excluded for further evaluation. In terms of invasion levels, nine melanomas were in situ melanomas (32%), and nine invasive melanomas were up to 1 mm thick (32%). A total of seven melanomas had a tumor thickness between 1 and 2 mm (25%), two melanomas were between 2 and 4 mm thick (7%), and one melanoma was thicker than 4 mm (4%). Regarding the subtype, 17 melanomas were classified as superficial spreading melanomas (61%), 2 were nodular (7%), 2 were acrolentiginous (7%), and 7 lentigo maligna (3) or invasive lentigo maligna melanomas (4) (25%). TNM classifications for the melanomas were T0 (9), IA (9), IB (7), IIA (1), IIB (1), and IIIC (1). All melanomas with a tumor thickness up to 2 mm were excised with a safety margin of 1 cm, while melanomas with a tumor thickness ≥ 2mm received excision with a safety margin of 2 cm. In patients with a melanoma with a tumor thickness > 1 mm, a sentinel lymph node biopsy (or in high-risk cases with a tumor thickness > 0.75 mm) was conducted.A total of 1 out of 36 lesions imaged with RCM and 4 from 84 lesions imaged with LC-OCT were classified as having poor image quality. The diagnostic confidence level depended on image quality for LC-OCT and RCM (Table 1, Figure 1). Lower image quality led to a poorer confidence level (Table 2). We found no difference between RCM and LC-OCT in terms of average image quality or average diagnostic confidence level (Table 2). A total of 77.8% of the lesions had good LC-OCT image quality (scored 0 or 1), while 72.2% had good RCM image quality. The average quality was evaluated as 1.2 for LC-OCT vs. 1.3. for RCM. However, the image quality (p = 0.49) and the confidence level (p = 0.40) were not significantly different between LC-OCT and RCM (p = 0.49). In total, 86.1% of the lesions had a high LC-OCT confidence level (scored 0 or 1), while 77.8% had a high RCM confidence level. The average confidence level was calculated as 1.2 for LC-OCT vs. 1.2 for RCM. Table 3 shows LC-OCT performances for diagnosing a melanoma vs. a nevus (dysplastic or not), and Table 4 for diagnosing a melanoma vs. a nevus vs. a dysplastic nevus for all 84 lesions. Figure 2 shows an example of a compound nevus, Figure 3 a dysplastic nevus and Figure 4 a melanoma with all three devices compared to histology. The case diagnosed as “other” (pigmented actinic keratosis) in histology was dropped for the evaluation. The accuracy of all LC-OCT performances for diagnosing a melanoma vs. nevus was 97.6%, the sensitivity was 92.9%, and the specificity was 100% (see Table 5). The specificity and sensitivity of LC-OCT for diagnosing a melanoma vs. a nevus were compared with the specificity and sensitivity of histology using McNemar’s test (p = 0.48). Two melanomas were falsely negative and were diagnosed as nevi with LC-OCT (for the analysis, see Table S5). Table 6, Table 7 and Table 8 illustrate the subgroup of 36 lesions that were imaged with LC-OCT and RCM. The accuracy of all performances with LC-OCT was 97.1% vs. 94.3% with RCM, the sensitivity for both was 92.9%, and the specificity with LC-OCT was 100% vs. 95.2% with RCM. The specificity and sensitivity of both devices for diagnosing a melanoma vs. a nevus were compared with the specificity and sensitivity of histology and with each other using McNemar’s test (for all p = 1). One false negative was found, which was the same lesion for LC-OCT and RCM, and which had a bad image quality score in LC-OCT (3) and in RCM (2). The most significant criteria for diagnosing a melanoma with LC-OCT were irregular honeycombed pattern (92% occurrence rate; 31.7 OR), presence of pagetoid spread (89 % occurrence rate; 23.6 OR) and absence of dermal nests (23 % occurrence rate, 0.02 OR) as seen in Table 9.Due to the high resolution and good penetration depth of LC-OCT, it is finally possible to overcome the gap between OCT and RCM. The vertical view is similar to OCT, and visualizes single cells, making the comparison with histology very intuitive. Monnier et al. already proved that LC-OCT can discriminate different skin levels and keratinocytes in healthy skin acquisitions [15,16,17]. Recent studies show that with single cell display, even BCC subtypes can be discriminated in vertical LC-OCT [18]. The disadvantage of LC-OCT is the lower penetration depth compared to OCT, so it is possible that deeper tumor parts may be missed [19,20]. Due to the presentation of single cells, differential diagnoses such as sebaceous hyperplasia, actinic keratoses, and squamous cell carcinomas could be differentiated, and the proliferation degree of actinic keratoses showed a 75% concordance between LC-OCT and histology [21,22,23,24]. This high single cell resolution is the reason why we assumed that even melanocytic lesions can be evaluated with LC-OCT. Regarding resolution, the horizontal LC-OCT images were very similar to RCM, which is the reference standard. With RCM it has already been shown that nevi can be discriminated from melanomas, even if there are sometimes possible false-negative and false-positive cases of melanomas [25]. Moreover, ex vivo RCM has also been successfully used for the differentiation and diagnosis of melanocytic lesions, although not all typical in vivo features could be detected since it offers a vertical view instead of horizontal as in in vivo RCM [26]. Hartmann et al. stated that ex vivo RCM might also be useful in the measurement of tumor thickness, and therefore might be of help for the presurgical definition of correct margins [27]. LC-OCT is similar to a fusion of ex vivo and in vivo RCM. It provides the vertical view of ex vivo RCM and the horizontal view of in vivo RCM on melanocytic lesions. Therefore, more information on pigmented lesions can be gained, and more typical features of both techniques can lead the clinician to the correct diagnosis. In our study, we found that image quality was responsible for the diagnostic confidence level for LC-OCT and RCM. In general, reduced image quality is mainly associated with ulceration, crusting, or image taking experience. Nevertheless, it is not recommended to measure an ulcerated or crusted pigmented lesion with non-invasive devices, and in such cases a biopsy is needed. Clearly, the diagnostic confidence for each lesion depends not only on the quality of the LC-OCT image or the difficulty of the lesion, but also on the experience of the observer. Our clinicians were all similarly experienced, and therefore we did not evaluate observer variability. Hence, further studies with clinicians of different experiences with LC-OCT need to be performed. We also conclude that for the interpretation and analysis of LC-OCT images, tele-consulting might be useful for LC-OCT beginners and for discussing difficult cases. Furthermore, the diagnostic performance of LC-OCT for melanomas (vs. all nevi) has the same sensitivity and a better specificity compared to RCM. We only had two false negative cases with LC-OCT, where two melanomas were diagnosed as nevi. We reviewed the cases, and in the first case—a nevus-associated melanoma—we detected in the integrated dermoscopic view that only the dysplastic nevus part had been imaged. Thus, it is very relevant to ensure appropriate coverage of the whole lesion, or if the lesion is too big, that more images are taken. The second case—a big in situ superficial spreading melanoma—was of a bad image quality, more in LC-OCT (score 3) than in RCM (score 2), because of an air bubble due to technical issues at that time. In big-sized lesions, multiple measurements from different parts should have been taken. Due to the lesion’s large size, a biopsy or surgery would have been performed anyway. Moreover, the diagnosis of a nevus vs. a dysplastic nevus is less accurate for both techniques. This aspect needs to be reconsidered, since one limitation of our study is that the number of dysplastic nevi in the subgroup of both devices was quite small (n = 3) vs. nevi (n = 19). A larger study for the evaluation of nevi vs. dysplastic nevi is required. One should keep in mind that histology can also be erroneous, especially in biopsies, because the entire lesion cannot be assessed here. Since the vast majority of the lesions were completely excised and serial sections including immunohistologies are standard in melanomas, we assume that this possible error is negligible. Recently Lenoir et al. published a few LC-OCT criteria for benign dermal melanocytic proliferations such as wave pattern [28]. In our study, we evaluated the most significant criteria for diagnosing a melanoma vs. nevus with LC-OCT (in comparison with RCM). We found that an irregular honeycombed pattern, the presence of pagetoid spread, and the absence of dermal nests are the most important criteria to discriminate a melanoma from a nevus in LC-OCT. These findings are very similar to studies with RCM [29,30,31]. We were surprised that no criteria related to the DEJ could be detected here, but this can be explained by the fact that many superficial spreading melanomas have a well-defined DEJ, which were the majority of the melanomas in our study. In the future, a study about thin versus thick melanomas should be conducted, since Rudnicka et al. showed in RCM that there might be different key criteria [32].In conclusion, our first study with the new LC-OCT device on melanocytic lesions showed that the discrimination between melanomas and nevi is possible. The improved distinction between nevi and melanomas with LC-OCT will lead to an immediate in vivo and non-invasive diagnosis, spare unnecessary surgeries if the lesion is diagnosed as a nevus, and therefore reduce health insurance costs and free capacities for other necessary melanoma surgeries.The following are available online at https://www.mdpi.com/article/10.3390/cancers14051140/s1: Table S1: Most common LC-OCT parameters for melanocytic lesions; Table S2: LC-OCT key criteria more useful in discriminating a melanoma from a nevus (dysplastic or not); Table S3: Most common RCM parameters for melanocytic lesions; Table S4: RCM key criteria more useful in discriminating a melanoma from a nevus (dysplastic or not); Table S5: Analysis of the false negative cases.Conceptualization, J.W. and E.C.S.; methodology, J.W., C.R., E.C.S. and S.S; validation, J.W., C.R., E.C.S. and S.S.; formal analysis, J.W., C.R., E.C.S. and S.S.; investigation, J.W., C.R., E.C.S. and S.S.; resources, J.W., C.R., E.C.S. and S.S.; data curation, C.R., C.G., F.D. and S.S.; writing—original draft preparation, S.S.; writing—review and editing, E.C.S., J.W., S.S., C.R., C.G., F.D. and M.K.E.P.; visualization, S.S.; supervision, E.C.S. and J.W.; project administration, J.W., E.C.S., C.R. and S.S.; funding acquisition, C.R. All authors have read and agreed to the published version of the manuscript.This study was partially funded by the FöFoLe, a Funding program for research and teaching and Research Grant of the medical faculty of the Ludwig Maximilian University of Munich (LMU Munich), protocol number 10-22.The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of LMU Munich (Protocol Number 17-699). Written informed consent was obtained from all subjects involved in the study. Fully anonymized data are available on request.The authors thank DAMAE medical, together with Maxime Cazalas and Mélanie Pedrazzani for providing the device needed for this study and for their constant professional assistance.The authors declare no relevant conflict of interest. Correlation between confidence level and image quality for LC-OCT and RCM. There was a statistically significant positive correlation between LC-OCT confidence levels and image quality for LC-OCT (rs(36) = 0.40; p = 0.02), as well as for RCM (rs(36) = 0.43; p = 0.008).(1) A compound nevus in comparison with vertical LC-OCT and OCT. A compound nevus on the left lumbal trunk. (a) Clinical, (b) dermoscopical, and (c) optical coherence tomography (OCT; 6 mm × 2 mm) images of a representative nevus of the study. (d) In vertical line-field optical coherence tomography (LC-OCT; 1.2 mm × 0.5 mm) images junctional and dermal nests, a well-defined dermo-epidermal junction (DEJ), and a papillomatous surface can be seen. (2). A compound nevus in comparison with horizontal LC-OCT and RCM. The same compound nevus is depicted in horizontal LC-OCT (1.2 mm × 0.5 mm) with a regular honeycomb pattern (a), with nests in the upper dermis and regular papillae (b). Reflectance confocal microscopy (RCM; 500 µm × 500 µm) shows the same features. A regular honeycomb pattern (c) and junctional nests, as well as regular papillae, but just a little bit brighter (d). (3). A compound nevus in histology. The histology shows the same compound nevus as in (1) and (2) with 4× magnification.(1). A dysplastic compound nevus in comparison with vertical LC-OCT and OCT. A dysplastic nevus on the left lower trunk. (a) Clinical, (b) dermoscopical, and (c) optical coherence tomography (OCT; 6 mm × 2 mm) images of a representative dysplastic nevus of the study. (d) In vertical line-field optical coherence tomography (LC-OCT; 1.2 mm × 0.5 mm) images, junctional and dermal nests, a well-defined dermo-epidermal junction (DEJ), but also a few bright atypical cells are visible. (2) A dysplastic compound nevus in comparison with horizontal LC-OCT and RCM. The same dysplastic compound nevus is depicted in horizontal LC-OCT (1.2 mm × 0.5 mm) with a regular honeycomb pattern (a), with nests in the upper dermis, less regular papillae, and a few bright atypical cells (red arrows) (b). Reflectance confocal microscopy (RCM; 500 µm × 500 µm) shows the same features. A regular honeycomb pattern (c) and junctional nests, less regular papillae, and also a few bright atypical cells (red arrows) (d). (3) A dysplastic compound nevus in histology. The histology shows the same compound nevus as in (1) and (2) with 4× magnification.(1) A melanoma in comparison with vertical LC-OCT and OCT. An ulcerated nodular melanoma on the upper right leg. (a) Clinical, (b) dermoscopical, and (c) optical coherence tomography (OCT; 6 mm × 2 mm) images of a representative melanoma of the study. (d) In vertical line-field optical coherence tomography (LC-OCT; 1.2 mm × 0.5 mm) images, bright atypical melanocytic cells (red arrows) and a disturbed dermo-epidermal junction (DEJ) are visible. No dermal nests can be seen. (2) A melanoma in comparison with horizontal LC-OCT and RCM. The same melanoma is depicted in horizontal LC-OCT (1.2 mm × 0.5 mm) with an irregular honeycomb pattern with single atypical cells (red arrows) (a), and a pagetoid spread with atypical melanocytes (red arrows) (b). There were no dermal nests, a more chaotic structure, and no edged papillae visible. Reflectance confocal microscopy (RCM; 500 µm × 500 µm) shows the same features. An irregular honeycomb pattern with single atypical cells, just brighter (red arrows), (c) and a lighter pagetoid spread with atypical melanocytes (red arrows) (d). There were also no dermal nests, a more chaotic structure, and no edged papillae visible. (3) A melanoma in histology. The histology shows the same ulcerated nodular melanoma as in (1) and (2) with a tumor thickness of 4.8 mm and with 4× magnification.Influence of image quality on the LC-OCT and RCM confidence level. LC-OCT and RCM image quality was scored from 0 (higher quality) to 3 (lower quality). LC-OCT and RCM confidence level was scored from 0 (higher confidence level) to 3 (lower confidence level).Comparison of image quality and confidence level between LC-OCT and RCM.LC-OCT performances for diagnosing melanomas vs. nevi (dysplastic or not) compared to histology.LC-OCT performances for diagnosing melanomas vs. nevi vs. dysplastic nevi compared to histology.LC-OCT performances for diagnosing melanomas vs. nevi (dysplastic or not).LC-OCT and RCM performances for diagnosing melanomas vs. nevi (dysplastic or not) compared to histology.LC-OCT and RCM performances for diagnosing melanomas vs. nevi vs. dysplastic nevi compared to histology.LC-OCT and RCM performances for diagnosing melanomas vs. nevi vs. dysplastic nevi.LC-OCT and RCM key criteria that were more useful in discriminating a melanoma from a nevus (dysplastic or not).Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01141.txt

@@ -0,0 +1 @@
+Esophageal cancer (EC) is the eighth most frequent cancer worldwide, with a poor prognosis. Current imaging modalities for staging and follow-up mainly include computed tomography (CT), positron emission tomography (PET)/CT, and endoscopic ultrasound. Magnetic resonance imaging (MRI), which is a non-irradiating and non-invasive modality, can provide identification of the esophageal wall and esophagogastric junction. MRI has shown encouraging capabilities in regional and local staging of EC as well as in the assessment of treatment response to therapy. Technical refinements in MRI technique and sequences overtime have contributed to its increasing diagnostic performance as well as its generalizability. MRI could become a routine imaging technique for EC management in the future, alone or in combination with other modalities.Esophageal cancer (EC) is the eighth more frequent cancer worldwide, with a poor prognosis. Initial staging is critical to decide on the best individual treatment approach. Current modalities for the assessment of EC are irradiating techniques, such as computed tomography (CT) and positron emission tomography/CT, or invasive techniques, such as digestive endoscopy and endoscopic ultrasound. Magnetic resonance imaging (MRI) is a non-invasive and non-irradiating imaging technique that provides high degrees of soft tissue contrast, with good depiction of the esophageal wall and the esophagogastric junction. Various sequences of MRI have shown good performance in initial tumor and lymph node staging in EC. Diffusion-weighted MRI has also demonstrated capabilities in the evaluation of tumor response to chemoradiotherapy. To date, there is not enough data to consider whole body MRI as a routine investigation for the detection of initial metastases or for prediction of distant recurrence. This narrative review summarizes the current knowledge on MRI for the management of EC.According to the global cancer observatory, esophageal cancer (EC) is currently the eighth most common cancer worldwide and the sixth most common cause of death by cancer (https://gco.iarc.fr/, accessed on 17 December 2021). About 600,000 cases were diagnosed in 2020, with more than 540,000 deaths. Both incidence and mortality rates vary greatly across countries [1]. Until recently, surgery and the combination of chemoradiotherapy have been the main treatments for EC in the localized setting. The recent CheckMate 577 trial has shown improved disease-free survival for patients treated with nivolumab, an immune checkpoint inhibitor, in the adjuvant setting of operated EC after neoadjuvant chemoradiotherapy [2]. In the metastatic setting, chemotherapy and immune checkpoint inhibitors are the main available treatments [3,4].Upon diagnosis, clinical and radiological staging is crucial to propose the optimal treatment strategy to patients. Several imaging techniques, including computed tomography (CT), positron emission tomography (PET)/CT, endoscopic ultrasound (EUS), as well as esophagogastroduodenoscopy, are recommended by European guidelines in the management of EC [5,6]. A preoperative CT classification has been proposed by Bosset et al. (cTNM) [7]. The TNM staging system, provided by the American Joint Committee on Cancer (AJCC) and the International Union Against Cancer (UICC), is used for pathologic tumor classification of the disease [8]. Contrary to other modalities, magnetic resonance imaging (MRI) is a non-irradiating and non-invasive technique. It also provides excellent soft-tissue contrast. It is not currently a routine examination for the management of EC because of its relatively low availability and its technical limitations. Nevertheless, it seems a promising technique for tumoral staging, delineation of target volumes before chemoradiotherapy, response to treatment and prediction of recurrence. Improvement of MRI modalities over the years, and the development of a larger choice of sequences, have contributed to enhancing MRI performance in EC.The purpose of this narrative review was to sum up the current evidence on the role of MRI in the management of EC.Measurements and description of the normal esophagus in the sagittal view were first assessed in 78 patients using electrocardiogram (ECG)-gated MRI. In 2004, Manabe et al. compared T1-weighted ultrafast gradient echo (TFE) MRI and T1-weighted fast field echo (FFE) MRI in 20 healthy volunteers, with the aim to delineate the esophageal passage under dynamic conditions [9]: The fast field echo images proved superior in terms of signal to noise ratio and overall quality [9]. The use of an external surface coil and cardiac gating with T2-weighted fast spin-echo (FSE) sequences was shown in 2006 to further improve the signal of the quality of esophageal MRI, and the speed of acquisition [10]. Later, an ex vivo study using high field MRI with a similar protocol helped precisely define the MRI anatomy of the posterior mediastinum [11]. Similarly, in a study on 33 operated patients with EC, preoperative high-resolution T2-weighted FSE sequence provided detailed images and comparison with histology findings showed good correlation between the degree of esophageal wall infiltration and pathological T-staging [12]. Furthermore, ultra-high-resolution T2-weighted MRI at 7.0-T provided clear definition of the esophageal wall and excellent accuracy for the T staging [13]. In 2017, pre-treatment motion-triggered MRI was used to improve the description of the periesophageal tissue [14].Dynamic MRI of the esophagus, using oral administration of various contrast agents, has been proposed to assess the esophageal peristalsis on sagittal images [15]. Gadopentetate dimeglumine mixed with barium [16], ferric ammonium citrate-cellulose paste [17], buttermilk spiked with gadolinium chelate [15], concentrated pineapple juice mixed with potato starch [18] were tested with satisfactory results, allowing to describe the esophagus in the sagittal plane over a mean of 16 cm, and define normal values for the esophageal transit time [19,20]. Noticeably, the latter intraluminal agent provided a similar image quality to those obtained with paramagnetic contrast agents.In our center, we obtain T1-weighted images in the coronal and axial planes before and after intravenous administration of a gadolinium chelate; then T2-weighted single shot spine echo sequences in the axial and coronal plane, and dynamic kinematic acquisition of steady-states sequences are obtained in the oblique plane, parallel to the esophagogastric junction. Patients are asked to swallow water through a straw during the dynamic acquisition phase. This is repeated several times in order to visualize the esophageal contractions. The sequences parameters are presented in the Table 1. The correct placement of sagittal oblique kinematic sequences is of paramount importance to visualize the esophagus and the esophagogastric junction. Indeed, the presence of the heart in the acquisition box will lead to motion artifacts. Morphological signs are analyzed on T2 weighed images and steady state cine sequences.We present in Figure 1 and Figure 2 the results of different modalities used for initial staging of EC in two different patients.One of the major challenges in EC is the evaluation of local staging in order to choose the best treatment approach. For low stage EC, the optimal treatment between upfront surgery or administration of neoadjuvant therapy is still unclear and will be influenced by robust staging.Tumor (T) staging is often done with EUS. In 2008, a meta-analysis on preoperative EC including 49 studies found pooled sensitivities ranging from 81.6 to 92.4% for differentiation of different T-stages, with a better performance in advanced (T4) disease [21]. Pooled specificities were 99.4% and 97.4% for T1 and T4 cancers, respectively [21]. In a more recent meta-analysis on preoperative ESCC, the overall accuracy of EUS for T-staging was 79% (95% CI: 88–94) [19]. EUS is superior to CT for evaluation of T-staging since CT cannot distinguish the different histological layers [20,22,23]. CT can be reliable when it comes to determining resectability by excluding high T-stages tumors [24,25]. 18F-fluorodeoxyglucose (18F-FDG) PET/CT has a limited role for T-staging due to its low spatial resolution and is mainly useful for the diagnosis of distant metastases [24,26]. Although the local and regional staging accuracy of EUS is greater than those of CT and PET [27], EUS is invasive and operator dependent, and sometimes limited by tumor stenosis [28].With this in mind, MRI seems a promising tool for the evaluation of T-staging in EC. Various in vitro studies found that MRI could clearly describe the different layers of the esophageal wall and had a high diagnostic accuracy for evaluating mural invasion [29,30]. Similar results were found in the ex-vivo setting with both high-resolution T2-weighted and diffusion-weighted MRI (DWI) [31,32,33], as well as in the in vivo setting [12,13]. Improvement of MRI modalities over the years has contributed in increasing the performance of MRI diagnosis and staging in EC [34,35]. Nevertheless, the heterogeneity of MRI modalities and sequences used in different studies limits the generalizability of currently available results. The T2-weighted FSE technique evaluated in 39 patients showed high accuracy in differentiating between T2 and T3 disease but with a tendency to overstage T1 tumors [12]. One trial evaluating the influence of two different volumetric interpolated breath-hold sequences (VIBE) on T-staging in EC, found that contrast-enhanced free-breathing radial VIBE was superior to breath-hold Cartesian VIBE, especially for T1 and T2 stage EC [36]. MR esophagography with water swallowing was evaluated in 30 patients with thoracic EC and 10 healthy volunteers [37]. By comparison with conventional MRI, it showed better results for assessing the tumor’s length and exact localization, but lower accuracy for T-staging [38]. T2*-weighted imaging had good accuracy for the evaluation of T-staging in patients with ESCC, except for the differentiation between T0 and T1-stage tumors [39]. In a study by Wu et al., gross tumor volume (GTV) assessed on T2-weighted MRI, contrast-enhanced T1-weighted and DWI in 60 patients with ESCC was associated with T-stage and the presence of lymph node metastases [40]. They also reported that GTV diagnosed on contrast-enhanced T1-weighted imaging better predicted T-stage [40]. Furthermore, one work suggested that whole-tumor histogram analysis of some pharmacokinetic parameters from dynamic contrast-enhanced (DCE)-MRI might be able to predict T-stage in ESCC [41]. In addition, MR angiography of the thorax in the same session provides very useful information about the invasiveness of the cancer in vessels, vascular anomalies, including arterial and venous status.A recently published meta-analysis on 20 trials addressed the issue of MRI diagnostic performance for EC, including the question of precise T-staging [35]. With 11 trials published between 2009 and 2019 addressing the question of differentiation between T0 and T1 or more advanced disease, MRI had a pooled sensitivity of 92% (95% CI: 82–96) and specificity of 67% (95% CI: 51–81). The administration of neoadjuvant chemoradiotherapy did not significantly impact these results [35]. With 10 studies evaluating differentiation between T2 or lower disease and T3 or higher disease, MRI had a pooled sensitivity of 86% (95% CI: 76–92) and a specificity of 86% (95% CI; 75–93). Unfortunately, the authors did not address the specific question of the diagnostic value of MRI to differentiate between T0 or T1 tumor (amenable to endoscopic low morbidity resection) and ≥T2 tumors (requiring chemoradiotherapy and/or surgical resection). Overall, MRI had a good sensitivity for T-staging in EC.Evidence in favor of a high accuracy for T-staging by MRI in preoperative EC is growing, even if there is heterogeneity between available studies in terms of MRI sequences, study designs and histological subtypes of EC. MRI shows good sensitivity for low T-stages and good sensitivity and specificity for higher T-stages. In the close future, MRI alone, or in combination with other modalities, will probably be used in routine clinical practice for early T-staging of EC.Upon diagnosis, regional staging with evaluation of node (N) staging is also crucial to evaluate prognosis and choose the optimal treatment approach. Not only is survival correlated with the T-stage, it is also clearly influenced by the N-stage [42]. Indeed, pathological evidence of lymph node metastases is a major prognostic factor in operated EC [37,42] with 5-year overall survival rates ranging between 70 and 92% in case of negative node involvement versus 18% to 47% for patients with positive lymph node involvement [43,44]. Furthermore, the lymph node ratio, which is the number of infiltrated lymph nodes divided by the total number of resected lymph nodes, is also an independent factor of survival for operated patients [45].Currently, baseline regional lymph node involvement in EC is also preferably evaluated with EUS, followed by CT and 18F-FDG-PET/CT [24,25,26]. EUS can give access to fine needle aspiration (FNA) for histologic evaluation of regional lymph nodes (mediastinum and coeliac). In a 2008 meta-analysis, EUS showed a pooled sensitivity of 80% (95% CI: 75–84) and pooled specificity of 70% (95% CI: 65–75) for N-staging [46]. Several studies have suggested that EUS-FNA had a greater accuracy than EUS alone for N-staging [47,48]. In the meta-analysis by Puli et al., pooled sensitivity of N-staging with EUS improved from 84.7 (95% CI: 82.9–86.4) to 96.7% (95% CI: 92.4–98.9) with FNA [21]. In the meta-analysis by Van Vliet et al., both CT and 18F-FDG-PET/CT showed lower sensitivities for regional N-staging in EC, 50% (95% CI: 41–60) and 57% (95% CI: 43–70), respectively [46]. Regarding 18F-FDG-PET/CT, this could partly be explained by the difficulty in the distinction of lymph nodes adjacent to a highly avid primary tumor with a high standard uptake value.Because of the previously described limitations of current techniques for N-staging, MRI has also been evaluated in this setting. Early studies using conventional MRI at 0.35–1.5 T without fast sequences reported sensitivities, specificities and diagnostic accuracies of 25–70%, 67–93% and 56–89% respectively [49,50,51,52]. More recent studies with similar modalities support these findings and found similar sensitivities, specificities, and accuracies of 38–62%, 68–85% and 64–77%, respectively [53,54]. Superparamagnetic iron oxide (SPIO), which is phagocytized by macrophages after intravenous administration, has enhanced the value of MRI in detecting lymph node metastases. Indeed, metastatic lymph nodes show a marked reduction in the uptake of SPIO due to a reduction in the number of phagocytes [55,56]. In a study on 16 patients by Nishimura et al., results were superior with an ultrasmall SPIO-enhanced MRI with sensitivity, specificity and accuracy of 100%, 95% and 96% respectively [54]. Limits to these results were the small number of patients included and the evaluation of differences between positive and negative lymph node groups rather than node-positive versus node-negative patients [54]. Later, a feasibility study on nine patients with EC and preoperative positive lymph node status, showed that ultrasmall SPIO-enhanced MRI could identify the majority of these suspected lymph node metastases [57]. In 2009, a study on 24 consecutive patients with EC showed that whole body DWI with background body signal suppression did not result in major diagnostic improvements for N-staging [58]. Recent evidence from a 2020 study on 76 patients suggested otherwise, with DWI showing higher sensitivity than 18F-FDG-PET/CT for the diagnosis of metastatic lymph node in ESCC [59]. In 35 patients with EC, ECG-triggered 1.5 T MRI with turbo spin-echo (TSE) and fast short tau inversion recovery (STIR) fat suppression yielded 81% sensitivity and 98% specificity for the diagnosis of lymph node involvement [60]. As mentioned previously, in the study by Wu et al., GTV assessed on T2-weighted imaging, contrast-enhanced T1-weighted and DWI in 60 patients with ESCC was associated with the presence of lymph node metastases [40]. Another study in 46 patients with EAC found similar results [61]. Recently, a radiomic signature with nine MRI features developed in a training cohort of 90 patients and confirmed in a validation cohort of 90 patients, showed good discrimination between metastatic and non-metastatic lymph nodes [62]. As for T-staging, one study suggested that whole-tumor histogram analysis of some pharmacokinetic parameters from dynamic contrast-enhanced MRI might be able to predict regional lymph node metastases in ESCC [41].Performance of MRI for lymph node assessment was also evaluated in the recently published meta-analysis by Lee et al. [35]. With 10 trials published between 2007 and 2019 addressing the question of differentiation of N0 disease from N1 and more advanced disease, MRI had a pooled sensitivity and specificity of 71% (95% CI: 60–80) and 72% (95% CI: 64–79) respectively [35].Again, here, improvement of MRI modalities over the years has positively influenced its diagnostic performance for N-staging in EC. Even if recent evidence is also in favor of good sensitivity and specificity of MRI for N-staging, it remains difficult to draw firm conclusions due to the heterogeneity of MRI techniques used in the different existing studies and their small sample sizes.Assessment of distant metastases in EC is currently done with CT and 18F-FDG-PET/CT. As previously mentioned, a 2004 meta-analysis evaluating the performance of 18F-FDG-PET/CT for the diagnosis of distant metastases found a pooled sensitivity and specificity of 67% (95% CI: 58–76) and 97% (95% CI: 90–100), respectively [26]. In the 2008 meta-analysis by Van Vliet at al., sensitivities and specificities for the diagnosis of distant metastases were 71% (95% CI: 62–79) and 93% (95% CI: 89–97) for 18F-FDG-PET/CT, 52% (95% CI: 33–71) and 91% (95% CI: 86–96) for CT [46]. In one study, restaging with 18F-FDG-PET/CT after neoadjuvant treatment of EC resulted in the diagnosis of metastases in 8% of patients, but also with false positive findings in 5% of patients [63]. Indeed, 18F-FDG-PET/CT shows several limitations that can affect its performance, such as the unspecific uptake of FDG, or the existence of low uptake tumors. Furthermore, 18F-FDG-PET/CT is an expensive and irradiating technique. For all these reasons, the used of 18F-FDG-PET/CT in EC is limited to the initial workup of ESCC, and this imaging modality is rarely used for the dynamic monitoring of tumors, CT being the preferred investigation.There is little data on the role of MRI in initial M-staging for EC. Two studies have evaluated the performance of whole-body MRI in that setting, one specifically in EC, and a second in a population with mixed gastrointestinal cancers including EC [64,65]. Compared with 18F-FDG-PET/CT, whole-body MRI had similar accuracy in detecting the primary tumor and lymph node metastases, and for excluding systemic metastatic disease [65]. In 49 patients with EC, both imaging modalities were able to identify distant metastases in two patients [65]. To date, there is however not enough data to recommend the routine use of whole-body MRI for M-staging in EC.Accurate tumor delineation before radiotherapy, including accurate GTV, is important to ensure adequate target coverage while limiting toxicity for surrounding organs at risks. MRI is already used for tumor delineation before chemoradiotherapy in various tumor sites. Currently, delineation of EC GTV is mainly based on the combined use of CT and 18F-FDG-PET/CT. Nevertheless, studies suggest that the correlation between tumor length assessment by CT and pathology is weak with a frequent overestimation of this measurement by CT [66,67]. EUS has been proposed for evaluation of the longitudinal extent of the tumor, but results are difficult to translate in the radiotherapy planning process [68]. The excellent soft-tissue contrast of MRI could also substantially help increase the accuracy of tumor delineation in this setting. One study evaluated GTV delineation by 10 observers in six EC patients with MRI compared with 18F-FDG-PET/CT [69]. The GTV appeared smaller on breath hold T2-weighted and DWI compared to 18F-FDG-PET/CT acquired during free-breathing, and the main variation was seen in the cranial caudal direction [69]. Combined DWI and T2-weighted MRI sequences in two tumors of the gastrointestinal junction reduced decreased caudal border variation [69]. However, MRI delineation did not reduce interobserver variability in this study, which could partly be explained by a lack of experience of contouring GTV in EC with this imaging modality. In another study including 42 patients with operated ESCC, DWI was the more accurate modality for the measurement of GTV compared with CT and T2-weighted MRI [66]. The difference in tumor length between CT and pathology was 3.6 mm, while the difference in length between DWI and pathology was as low as 0.5 mm [66].Finally, MRI-guided radiotherapy for EC remains under development but seems a promising option for the future. The combination of improved GTV delineation, respiratory gating, and online adaptive planning from daily non-irradiating MRI, could allow for tighter target coverage while sparing the adjacent normal organs [70].Several studies have shown improved overall survival with neoadjuvant treatment in patients with localized EC compared with upfront surgery [71,72,73]. Nevertheless, after neoadjuvant treatment and surgery, about one third of patients with EC show complete pathologic response and are possibly being unnecessarily exposed to the risks of esophagectomy [73]. Moreover, some patients will not respond to neoadjuvant treatment but will be exposed to its side effects. Overall, there is a double challenge of early identification of non-responders to neoadjuvant treatment and precise restaging after chemoradiotherapy to detect residual disease.Previous studies and meta-analysis have shown that EUS, CT and 18F-FDG-PET/CT are not always adequate for the detection of residual disease in EC after neoadjuvant treatment, with poor overall accuracies [34,74,75,76]. A meta-analysis evaluating pathologic complete response after neoadjuvant treatment with various imaging techniques found pooled sensitivities of 35% (95% CI: 16–60), 62% (95% CI: 50–73), 1% (95% CI: 0–92), and 80% (95% CI: 46–95), and pooled specificities of 83% (95% CI: 71–91), 73% (95% CI: 64–81), 99% (95% CI: 81–100), and 83% (95% CI: 65–93) for CT, 18F-FDG-PET/CT, EUS, and MRI respectively [77]. Indeed, three-dimensional-CT volumetry evolution is not associated with histopathological tumor response [78]. EUS is limited by the difficulty of differentiating residual tumor from inflammation and fibrosis [79,80]. Finally, non-specific glucose uptake after inflammation, or the existence of low FDG uptake cancers limits the role of 18F-FDG-PET/CT in treatment response monitoring.DWI and the derived apparent diffusion coefficient (ADC) can help assess tumoral metabolic activity and have shown promising results for response prediction in EC in various studies [81,82,83,84,85]. Indeed, results suggest that changes observed between baseline diffusion-weighted images and interim diffusion-weighted images (during treatment) are good prognostic and predictive biomarkers. The relative change in ADC during the first two weeks of chemoradiotherapy appears to be the most predictive for the detection of residual cancer, with a sensitivity of 100% and specificity of 75% [83]. Moreover, various b values for ADC have been evaluated in the different available studies overtime [86]. One study has suggested promising results for the prediction of treatment response in EC with the use of intravoxel incoherent motion diffusion-weighted images, which can simultaneously obtain diffusion and perfusion information from tissues without administration of a contrast agent [87]. In addition to DWI, dynamic contrast-enhanced-MRI can also be used to predict response to chemoradiotherapy in EC [88,89]. Finally, weekly T2-weighted MRI in 29 patients undergoing neoadjuvant chemoradiotherapy was able to identify volumetric changes with a significant decrease in tumor regression volume overtime [90].In a recent meta-analysis including seven studies with EC patients treated by chemoradiotherapy, the pooled sensitivity and specificity of DWI for predicting early response to treatment were 93% (95% CI: 77–98%) and 85% (95% CI: 72–73) for the ΔADC (difference in ADC values before and after chemoradiotherapy) and 75% (95% CI: 62–84) and 90% (95% CI: 67–97) for the post ADC [91]. Even if included studies were heterogeneous with small sample sizes, these results suggest a role for DWI in the assessment of treatment response for EC.The ongoing SANO-2 trial (NCT04886635) is currently evaluating the role of active surveillance with 18F-FDG-PET/CT and endoscopic biopsies after neoadjuvant chemoradiotherapy. Similarly, the randomized ESOSTRATE trial (NCT02551458) is comparing active surveillance versus surgery in patients with compete pathological response after chemoradiotherapy. Finally, the ongoing prospective study PRIDE (NCT03474341) is evaluating a multimodal prediction model including DWI and dynamic contrast-enhanced MRI to predict patient’s individual probability of complete pathological response after neoadjuvant chemoradiotherapy and identify early non-responders [92].Recent improvement in technical modalities of MRI has allowed to better assess the morphology of the normal and pathologic esophagus. Available data in EC is still scarce with small sample size studies and heterogeneity regarding clinical setting and MRI sequences and modalities. Nevertheless, currently used techniques for EC management (CT, 18F-FDG-PET/CT and EUS) show significant limitations, making MRI a promising tool in both initial staging (T-staging and N-staging) and assessment of response to chemoradiotherapy. There is still not enough data to conclude on the potential role of MRI in detection of distant metastasis as well as follow-up.High field (7 T) MRI of the esophagus, currently only studied ex vivo, demonstrates an excellent sensitivity and specificity for esophageal cancer. Most importantly, it provides a clear image of the tissue layers of the esophageal wall, comparable to that of endosonography or pathology. The clinical use of this imaging modality—currently limited to the brain and the joints—for the workup of EC could allow an accurate noninvasive tumor staging, even distinguishing shallow T1 lesions potentially amenable to endoscopic resection from T1 lesions with deep submucosal infiltration or T2 lesions requiring surgical resection. In addition, its ability to discriminate fibrosis from neoplastic tissue makes of high-field MRI a promising candidate to better assess tumor response to neoadjuvant therapy.In EC, combinations of different diagnostic modalities might be the way to go for optimal individual staging. In a study on 19 patients with resectable EC, PET-MRI demonstrated acceptable accuracy for T-staging compared with EUS and, although not statistically significant, higher accuracy than EUS and 18F-FDG-PET/CT for prediction of N-staging [93]. Furthermore, radiomics and the use of various artificial intelligence-based systems using CT imaging have recently shown promising results in a variety of diseases, including the diagnosis and monitoring of EC [94,95,96]. It may be assumed that radiomics and artificial intelligence-centered studies in EC will probably consider data from MRI for even higher performance and accuracy.Conceptualization, M.B., A.D., P.S.; methodology, M.B.; validation, A.P., A.D., P.S., J.V., R.C., M.B.; formal analysis, A.P.; investigation, A.P.; resources, A.D., P.S.; data curation, A.P.; writing—original draft preparation, A.P.; writing—review and editing, A.P., A.D., P.S., J.V., R.C., M.B.; visualization, A.D.; supervision, M.B. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The authors declare no conflict of interest.Initial staging for a 66-year-old woman with T2N0M0 esophageal adenocarcinoma: (A) endoscopic view of an elevated neoplastic lesion arising on a Barrett’s esophagus on the left posterior side of the esophagus, 34 cm from the dental arch (arrow). (B) Endoscopic ultrasound showing the hypoechoic, well-limited neoplastic lesion in close contact with the muscularis propria without regional lymph nodes (arrow). (C) Contrast-enhanced computed tomography image obtained at 70 s after intravenous administration of iodinated contrast material and without oral contract material in the axial plane shows irregular esophagus (arrow) but no definite lesion. (D) Axial T2-weighted single shot magnetic resonance (MR) images. after oral administration of water show a low signal irregular anterior lesion (arrow). (E) Coronal contrast-enhanced T1 weighted images confirm the lesion that demonstrates an early contrast uptake (arrow). (F) Sagittal steady-state MR images confirm the low signal lesion (arrow) and allow to precisely see its location.Initial staging for a 77-year-old man with T3N1M0 esophageal adenocarcinoma. (A) Esophagram shows esophageal tumor of the lower third of the esophagus responsible for marked luminal narrowing (arrow). (B) T2-weighted HASTE MR image in the axial plane shows esophageal tumor (arrow) with luminal narrowing (arrowhead). (C) Diffusion-weighted MR image in the axial plane obtained with high b value (b = 800 s/mm2) shows restricted diffusion (arrow) consistent with malignant esophageal tumor. Additional hyperintense lymph node is present (arrowhead). (D) T1-weighted VIBE image in the axial plane obtained 30 s after intravenous administration of a gadolinium-based contrast agent (gadoterate meglumine, Dotarem®, Guerbet, Villepinte, France) shows heterogeneous esophageal tumor (arrow) and enhancing lymph node (arrowhead). (E) T1-weighted VIBE image in the axial plane obtained 60 s after intravenous administration of a gadolinium-based contrast agent (gadoterate meglumine, Dotarem®, Guerbet) shows that the tumor is well delineated without spreading outside the adventitia (arrow). (F) MR angiography image in the oblique plane shows intact interface (arrows) between esophageal tumor and aorta.How we do it in our center: different magnetic resonance imaging sequence parameters at 1.5 T (Siemens Aera, vb20a).TSE: turbo spin echo, EPI: echo planar imaging, TE: echo time, TR: repetition time, FOV: field of view, NA: not applicable.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01142.txt

@@ -0,0 +1 @@
+Prostate cancer is a major cause of health loss and death worldwide, and better tools to assess risk levels in individual patients are needed. MicroRNAs (miRNAs) are small molecules with critical regulatory roles in cell functions and are also involved in prostate cancer development. The aim for this study was to investigate the role of miR-24-1-5p regarding prognosis in men diagnosed with prostate cancer and treated with radical prostatectomy. We collected prostate cancer tissue from 142 men already enrolled in a population-based cohort study who underwent prostatectomy. We examined the tissue expression of miR-24-1-5p in prostate cancer using in situ hybridization (ISH) and semi-quantitative scoring. We found that a high miR-24-1-5p expression was associated with a doubled risk of recurrence of prostate cancer.The role of miR-24-1-5p and its prognostic implications associated with prostate cancer are mainly unknown. In a population-based cohort, the Prostate Cancer Study throughout life (PROCA-life), all men had a general health examination at study entry and were followed between 1994 and 2016. Patients with available tissue samples after a prostatectomy with curative intent were identified (n = 189). The tissue expression of miR-24-1-5p in prostate cancer was examined by in situ hybridization (ISH) in tissue microarray (TMA) blocks by semi-quantitative scoring by two independent investigators. Multivariable Cox regression models were used to study the associations between miR-24-1-5p expression and prostate cancer recurrence. The prostate cancer patients had a median age of 65.0 years (range 47–75 years). The Cancer of the Prostate Risk Assessment Postsurgical Score, International Society of Urological Pathology grade group, and European Association of Urology Risk group were all significant prognostic factors for five-year recurrence-free survival (p < 0.001). Prostate cancer patients with a high miR-24-1-5p expression (≥1.57) in the tissue had a doubled risk of recurrence compared to patients with low expression (HR 1.99, 95% CI 1.13–3.51). Our study suggests that a high expression of miR-24-1-5p is associated with an increased risk of recurrence of prostate cancer after radical prostatectomy, which points to the potential diagnostic and therapeutic value of detecting miR-24-1-5p in prostate cancer cases.Prostate cancer (PCa) is a major cause of health loss and death worldwide, and it is a heterogeneous disease [1,2]. Compared with localized low-risk PCa that can be actively surveyed without management, the treatment for aggressive high-risk PCa is most often systemic and complex. We need valid prognostic biomarkers to distinguish low-risk indolent PCa from aggressive PCa.MicroRNAs (miRNAs) are a class of endogenous non-coding small RNA molecules associated with the regulation of gene expression and are “fine-tuners” of the immune system [3]. These have been studied for their potential to serve as molecular prognostic biomarkers for cancer including PCa [4]. In particular, differential miRNAs’ expression profiles between tumour and normal tissues have been observed for PCa, as well as for other other cancer types [4,5]. In a recent systematic review, fifteen miRNAs were associated with PCa prognosis [4]. These are transcribed as ~70 nucleotide precursors in a stem-loop sequence and are subsequently processed by the Dicer enzyme to give two mature ~22 nucleotide products. These miRNAs bind to the 3/-untranslated region (3/-UTR) of target messenger RNA (mRNA) and are used to identify target mRNA transcripts. They can prevent protein expression through cleavage of specific target mRNAs or through inhibition of their translation, and thus influence developmental processes, tissue housekeeping and tumorigenesis [6]. Aberrant expression or dysregulation of miRNA can influence the activity of tumor suppressors or oncogenes in many human cancers [6,7], including prostate cancer [8]. An example of this is how miR-21 expression can trigger an epithelial to mesenchymal transition in aggressive prostate cancer cells through the Wnt signaling axis [9].Additionally, miRNAs have also been associated with the tumor microenvironment, as well as PD-L1 and STAT3 signaling in prostate cancer cells, supporting the idea that miRNAs play a role in and are linked to inflammation [10]. Most prostate tumors contain immune cells, and chronic inflammation, one of the hallmarks of cancer development [11,12], has been proposed as a key factor in prostate cancer development [13,14,15]. The suggested hypothesis is partly based on observations of inflammatory cells in the prostate microenvironment of adult men and partly by the observation that this inflammation has been associated with precursor lesions in the prostate gland, termed proliferative inflammatory atrophy [16,17,18,19]. However, much remains unknown regarding possible biological mechanisms operating in relation to prostate cancer development and systemic and local inflammation, and only several mechanisms, including miRNAs and factors related to the immune system, have been studied [3,20].The effects of miRNAs in prostate cancer have been studied, but the biological mechanisms operating, as well as the types of miRNAs and their functions, have not yet been clarified [4,6,21,22]. Importantly, no prostate-specific miRNAs have yet been definitively identified. We previously studied the association between several miRNAs and prostate cancer recurrence and survival [23,24,25,26,27,28]. High expressions of miR-205, miR-17-5p, miR-20a-5p, miR-210, and miR-141 and a low expression of miR-424 were all associated with an increased risk of prostate cancer recurrence. These miRNAs have been suggested to be associated with inflammation; however, there is limited knowledge [3]. Furthermore, few have investigated the association between miR-24 and prostate cancer [8]. Through deep sequencing of prostatectomy specimens, it was observed that miR-24 was downregulated compared to non-cancer prostate tissue [29]. Another study, by Hashimoto et al. found that miR-24 was differentially expressed in African American and Caucasian American prostate cancer patients [30]. Interestingly, miR-24-3p enhanced Paclitaxel sensitivity in Paclitaxel-resistant prostate cancer cells [31], while in xenograft cell lines, miR-24 was downregulated in metastatic prostate cancer compared to non-metastatic [32]. Furthermore, the miR-24 expression was significantly lower in prostate cancer cell lines compared to a normal prostate epithelial cell line. These findings suggest that miR-24 plays a tumor suppressor role in prostate cancer and targets p27 and p16 in prostate cancer cells [33]. Current knowledge about miR-24 is largely based on in vitro studies and/or mouse models. The stem-loop sequence hsa-miR-24-1 is the processor of two mature sequences: hsa-miR-24-1-5p and hsa-miR-24-3p [34]. To our knowledge, previous studies have not reported which sequences of miR-24 they have used [32,33].The present study is based on men participating in the Tromsø Study, a population-based cohort study, which has a high attendance proportion and long follow-up time [35]. Complete information on prostate cancer cases, including detailed medical and pathological records, has been obtained in a substudy, the Prostate Cancer Study throughout life (PROCA-life) [36]. The role of miR-24s, including the different types of miR-24 and their prognostic implications, is still under debate, and their potential diagnostic and therapeutic values are not clarified. Therefore, the main aim of the present study was to analyze the influence of miR-24-1-5p regarding aggressiveness and prognosis in men diagnosed with prostate cancer and treated with radical prostatectomy.The present study cohort, PROCA-life study, is based solely on men aged ≥ 25 years who were enrolled in the population-based Tromsø Study from 1994 to 2016 (Tromsø 4, 1994–95, Tromsø 5, 2001, Tromsø 6, 2007–2008, Tromsø 7, 2015–2016) [37]. The procedures for invitations, screening, and examinations were almost identical in all three surveys. Moreover, all data collection was performed by trained research technicians at one study site. Age-eligible men were invited to participate by a personal invitation [35,37]. A total 75.6% of invited men attended, completed questionnaires, and provided biological specimen samples and clinical measurements.Height and weight were measured on an electronic scale with the participants wearing light clothing and no shoes. Height was measured to the nearest 1 cm (cm) in Tromsø 4 and nearest 0.1 cm in Tromsø 5–7. Weight was measured to the nearest 500 g in Tromsø 4 and nearest 100 g in Tromsø 5–7. Body mass Index (BMI) was calculated using the formula weight/height2 (kg/m2). Blood pressure (BP) was measured on the right arm three times at one-minute intervals after two minutes of seated rest, and the mean of the last two measurements was used. Information about lifestyle factors was obtained from the questionnaires. Alcohol consumption was defined as more than 1 unit (drink) of alcohol per month, as described by others in the same cohort [38,39].Blood samples were drawn by trained research assistants on attendance at each survey and were non-fasting. Analyses of serum samples were completed at the Department of Laboratory Medicine, University Hospital of Northern Norway (UNN), Tromsø, Norway [35]. For white blood cell count (WBC), 5 mL of blood was collected into Vacutainer tubes containing K3-EDTA 40 lL, 0.37 mol/L per tube, and analyzed within 12 h by an automated blood cell counter (Coulter CounterÒ and Coulter LH750 Coulter Electronics, Luton, UK). Total cholesterol and triglyceride levels were analyzed by enzymatic colorimetric methods with commercially available kits (CHOD-PAP for cholesterol). Prostate Specific Antigen (PSA) measurements were taken for prostate cancer cases only, as a part of the clinical routine for diagnosis and follow-up (1990–1994 Stratus® PSA Fluorometric Enzyme Immunoassay, (BDI, Miami, FL, USA), 1994–2001 AxSYM Psa Reagent Pack, (Abbott®, Lake Bluff, IL, USA), 2001–2020 Bayer® PSA Reagent Pack Immuno I (Prod. Nr. T01-3450-51, Technicon Immuno I (New York, NY, USA).Prostate cancer cases during follow-up (until 31 December 2018) were identified in the Cancer Registry of Norway (n = 947), by using the  unique national 11-digit identification numbers (The National Population Registry at Statistics Norway). Cases with available tissue samples after prostatectomy with curative intent were identified by cross-linkage with the archive of Department of Clinical Pathology, University Hospital of North Norway, Tromsø, Norway (n = 189), and these constituted the eligible study population in the current study (see flow chart figure in Appendix B). Overall, 43 cases were not technically successful in the in situ hybridization (ISH) staining process and were excluded. Furthermore, four cases were excluded because they did not have curative surgery, leaving a final study population of 142 men (Figure A1).Detailed clinical information was obtained by trained physicians (MS, TK, and ES) and included prostate cancer treatments and recurrence. Cause of death was obtained through linkage with the Norwegian Death Registry by use of the unique personal identification number. Most of the prostate cancer patients (88.7%) underwent prostatectomy a few months after being diagnosed; the remainder of the study population (11.2%) underwent active surveillance until their prostate cancer showed signs of increasing aggressiveness. Date of prostatectomy was used for calculation of age and follow-up time. The current study is based on the Tromsø Study survey closest to the date of prostatectomy for baseline data such as height, weight, blood pressure, triglyceride levels, and alcohol use.Histopathological information was obtained from medical records, but all histopathological specimens were re-examined by one specialized uropathologist (ER) and classified according to the latest International Society of Urological Pathology (ISUP) guidelines using their Gleason scores and ISUP grade groups [40]. Prostate cancer cases were divided into three risk groups based on PSA level at diagnosis, highest ISUP grade group, and clinical T-stage, according to classification guidelines from the European Association of Urology (EAU) [41]. Risk group 1 (low) was defined as: PSA <10 µg/L, clinical T-stage (cT-) 1, and ISUP grade group 1. Risk group 2 (intermediate) was defined as: PSA: 10–20 µg/L, cT-stage 2, or ISUP grade group 2–3. Risk group 3 (high) was defined as: PSA: >20–100 µg/L, cT-stage 3, or ISUP grade group 4–5. ISUP grade groups were reported after reclassification when available. Cancer of the Prostate Risk Assessment Postsurgical Score (CAPRA-S Score), a validated score developed to predict outcomes after radical prostatectomy, was also used to classify patients into risk groups [42]. This score is based on surgical margin, seminal vesicle invasion, extracapsular extension, lymph node invasion, PSA value, and Gleason/ISUP Grade Groups.Tissue microarrays (TMAs) were constructed for the analysis of ISH staining expression. For each case, one uropathologist (ER) identified and marked representative areas of the prostate specimens with tumor epithelial cells (TE) and normal epithelial cells (NE). From each of these areas, 0.6 mm cores were sampled from each donor block and inserted into paraffin blocks to construct TMA blocks by using a tissue-arraying instrument (Beecher Instruments, Silver Springs, MD, USA). The details of the technique were described earlier [43].The tissue expression of mature miR-24-1-5p in prostate cancer was examined by in situ hybridization (ISH). The principle of the method is based on the ability of specific microRNA locked nucleic acid (LNA) probes to bind to target microRNA in tissue followed by chromogenic visualization. ISH staining was performed automatically in a Ventana Discovery Ultra instrument. Necessary efforts to avoid RNA degradation in tissue were made in work routines and by using RNAse-free buffers during the process.LNA probe concentrations, hybridization temperatures, and incubation times were optimized before staining the tissue of interest. Target retrieval treatment was adjusted to improve availability of microRNA sequence for the target and control probes. A TMA multi organ block with several normal and tumor tissues was used for optimization of the ISH method and validation of miR-24-1 expression in different tissues. We used a U6snRNA probe as a positive control and to ensure the sensitivity level of the method. Strong nuclear U6snRNA staining also indicates a low degree of RNA degradation of the tissue. A scramble miRNA negative control probe showed no unspecific staining. Optimized ISH parameters are presented in Table A1, and details of the products are ordered in Table A2.External validation of the LNA probes was completed by supplier company QIAGEN. The LNA miRNA probes were purified by HPLC (High-Performance Liquid Chromatography) and analyzed by Capillary Electrophoresis or HPLC. The identity of compounds was confirmed by using Mass Spectrometry. For more details on the ISH procedure, see Appendix A.The expression of miR-24-1-5p was assessed by semi-quantitative scoring by two trained independent investigators (ES, ER). The color intensity was graded as negative (0), weak (1), moderate (2), strong (3), or missing (4) (Figure 1). Two areas of TE cells and two areas of NE cells were scored for each patient. The same methodology has been used by our group previously [24,28]. Stromal areas were not scored due to little positivity. Mean and median scores were calculated for TE and for NE separately, as well as for TE+NE combined. High expression of miR-24-1-5p was defined as a score equal to or higher than the median score of the study population. Inter-observer variability was assessed by calculating linearly weighted Kappa statistics and showed a moderate agreement (Kappa 0.59 (SD 0.50–0.68)).The primary endpoint was defined as a composite endpoint, including any evidence of recurrent prostate cancer after surgery, biochemical failure (PSA-level ≥0.2), and/or clinical/radiological signs of prostate cancer defined by the treating physician. Endpoints were updated until August 2021.Selected characteristics that describe the study population are presented as a mean (standard deviation), median (range) or percent (numbers). Spearman’s correlation coefficient was used for correlation analysis between miR-24-1-5p and clinicopathological markers. The five-year recurrence-free percentage was calculated using the Kaplan–Meier survival function, and statistical differences between different groups (e.g., ISUP grade group, EAU risk group, CAPRA-S) were tested by using log-rank test.Multivariable Cox proportional hazard models, with time after surgery as timescale, were used to study whether miR-24-1-5p and clinicopathological markers were independently associated with a risk of prostate cancer recurrence. Several variables were assessed as potential confounders based on suggested biological mechanisms and/or significant associations in unadjusted models. Age at surgery (continuous), CAPRA-S (categorical), BMI (continuous), alcohol habits (categorical), and cholesterol levels (continuous) were included in the final models as covariates. We performed a stratified analysis via systolic blood pressure based on previous observations suggesting that elevated systolic blood pressure is associated with prostate cancer risk [44]. The proportional hazard assumption was assessed by visually controlling that the log minus log survival curves were parallel. The Kaplan-Meyer method was used for drawing survival plots for high vs. low expression of miR-24-1-5p. We conducted all statistical tests with STATA/MP version 16 (StataCorp LLC, College station, TX, USA) and used a two-sided significance level of p < 0.05.The 142 men that constituted the study population entered the Tromsø Study on average 8.0 years before prostatectomy. The median age at prostate cancer diagnosis was 64 years (range 46–74 years), the median age at prostatectomy was 65 years (47–75 years), and prostatectomy was performed between 2001 and 2018 (Table 1). The prostate cancer patients had an average BMI of 27.1 kg/m2, systolic BP of 134.9 mmHg (SD 16.8), and diastolic BP of 80.4 mmHg (SD 9.4) at study entry. A total of 61.3% of the prostate cancer patients had a systolic blood pressure higher than 130 mmHg. The mean level of white blood cells was 6.60 × 109/L (SD 1.67), total cholesterol was 5.78 mmol/L (SD 1.12), and triglyceride level was 1.70 mmol/L (SD 0.90). Additionally, 46.1% were alcohol users.The surgical technique changed during the study period: 47.2% of the patients had open (retropubic or perineal) prostatectomy, mostly before the year 2012, while 52.8% had laparoscopic prostatectomy (manual or robot-assisted). Lymph node dissection was performed in 36.6% of the patients. The mean PSA at prostate cancer diagnosis was 10.5 ng/mL (SD 9.5). The histopathologic tumor stage was pT2c for 47.9% of the patients, while 26.1% had pT3, and the ISUP grade group was 1 or 2 for 73.8% of the patients. The median CAPRA-S score was 3 (39.4% Capra-S low (0–2), 46.5% Capra-S intermediate (3–5), and 14.1% Capra-S high (6–12). Positive surgical margins were found in 30.5% of the cases. Overall, 26.9% of the prostate cancer patients had a relapse after prostatectomy during follow-up (until August 2021).The mean score for miR-24-1-5p expression was 1.60 in TE cells, 1.35 in NE cells, and 1.49 in TE and NE cells combined (Table 2). The median value was used as cut-off value for high miR-24-1-5p expression and was ≥1.67 in TE and ≥1.50 in NE. The cut-off value for high TE+NE combined was ≥1.57. In the total population, 43.7% had high TE+NE, 43.7% had high TE, and 45.1% had high NE.The level of white blood cells at study entry (pre-diagnostic) correlated with miR-24-1-5p expression in both TE and NE (r = 0.21, p = 0.02 and r = −0.21, p = 0.01, respectively). Furthermore, BMI and triglyceride levels at study entry correlated with miR-24-1-5p expression in NE (r = −0.27, p = 0.01 and r = −0.24, p = 0.006). Positive surgical margin correlated with miR-24-1-5p expression in TE (r = 0.19, p = 0.029). CAPRA-S correlated with miR-24-1-5p expression in TE (r = 0.21, p = 0.020) (results not presented in table). There were no correlations between miR-24-1-5p expression and PSA at diagnosis, Gleason score, perineural invasion, age at surgery, or BMI.Age at surgery was not associated with recurrence-free survival (Table 3). Increasing CAPRA-S score, ISUP grade group, and EAU risk group were all significant prognostic factors for decreasing five-year recurrence-free survival (p < 0.001). Our data suggested a higher number of recurrences in the group with high expression of miR-24-1-5p but of borderline significance (p = 0.098) (Figure 2). In the subgroup of prostate cancer patients with high pre-diagnostic systolic blood pressure (≥130 mmHg), high expression of miR-24-1-5p was a prognostic factor for recurrence.In our multivariable model, we adjusted for age, Capra-S group, BMI, cholesterol level, and alcohol use, based on suggested biological mechanisms. High miR-24 expression in the tissue (TE + NE) was associated with an almost doubled risk of the recurrence of prostate cancer compared to that with low miR-24-1-5p expression (HR 1.99, 95% CI 1.13–3.51) (Table 4). The results were also observed in the subgroup of prostate cancer patients with high pre-diagnostic systolic blood pressure. There was no significant interaction between miR-24 expression and blood pressure, nor between miR-24 expression and follow-up time.We found that high expression of miR-24-1-5p was associated with an almost doubled risk of recurrence (biochemical or clinical) after radical prostatectomy, when adjusting for known histopathological risk factors. We were also able to adjust for known lifestyle risk factors due to the pre-diagnostic information assessed at the study entry. We did not observe correlations between age, perineural infiltration, PSA values, or Gleason score and miR-24-1-5p expression, however it did correlate with CAPRA-S, a score that incorporates the PSA and Gleason score. Of note, we observed positive surgical margins correlated with miR-24-1-5p expression.Prostate cancer is a heterogeneous condition, ranging from indolent to life-threatening, and we need better tools for disease stratification. Development of biomarkers for risk stratification, personalized treatment, and follow-up is needed. Other miRNAs have shown good correlation between levels in tissue and in blood or urine, and the development of liquid biomarkers would be a great advantage for the patient by limiting the need for invasive tissue biopsies. In addition, miRNAs possibly play a role in prostate cancer metastasis and are therefore potential targets for new therapeutic agents [45].To our knowledge, this is the first study to investigate whether the expression of miR-24-1-5p in prostate cancer tissue is associated with prognosis. Our findings are in part supported by others, although few studies have investigated the role of miR-24-1-5p in prostate cancer. Most of these studies have been experimental. A recent meta-analysis studied the prognostic significance of miR-24 in various cancers and found that high miR-24 expression was associated with poor overall survival [46]. The meta-analysis consisted of 17 studies, and a total of 1705 patients, of whom none had prostate cancer. Another recent study observed that the expression of miR-24-1-5p decreased 16-fold after radiotherapy doses of 6 and 7 Gy in prostate cancer cell lines treated with radiation, suggesting that expression of miR-24-1-5p may impact the efficacy of important treatment modalities, such as radiation therapy [47]. Further studies are needed to explore the importance of this observation.A few studies have evaluated the other mature sequence of miR-24, miR-24-3p, which has been suggested as a diagnostic biomarker for prostate cancer in serum [48,49]. The circRNA protein kinase C-iota has been suggested to influence tumor development, and a study found this molecule triggers growth and metastasis in prostate cancer by downregulation of miR-24-3p [50]. ACVR1B, BCL2, BIM, eNOS, FGFR3, JPH2, MEN1, MYC, p16, and ST7L are miR-24 targets that have been experimentally validated in human cells [20]. However, it is unclear whether these results will be valid for the association between miR-24-1-5p and prostate cancer development.The relationship between prostate cancer and inflammation has been the subject of several studies. Inflammation is one of the classic hallmarks of cancer [11], and inflammatory cells associated with the precursor lesions for prostate cancer in the prostate gland have been observed [13]. We have previously discovered that systemic pre-diagnostic inflammatory biomarkers were associated with prostate cancer development [36]. In mouse models, prostate-specific PTen deletion has been found to activate inflammatory microRNA expression pathways [51]. Additionally, miR-24 has been linked to inflammation [3]: miR-24 was found to regulate phagocytosis in myeloid inflammatory cells [52]. In a murine model, miR-24 was a central regulator of vascular inflammation [53]. In a model with primary human macrophages, miR-24 would produce anti-inflammatory action by inhibiting the production of pro-inflammatory cytokines, and these results suggest that overexpression of miR-24 would have mostly anti-inflammatory effects [54]. Conversely, miR-24 belongs to the miR-23~27~24 cluster, and this cluster has been shown to reduce TNF-α and IL-6 production [55]. In summary, the current literature on miR-24 is not consistent on whether miR-24 has a pro- or anti-inflammatory role. Our observation that the association between miR-24-1-5p and prostate cancer recurrence was suggestively more pronounced among the prostate cancer patients with high pre-diagnostic systolic blood pressure supports the possibility of a role associated with low-grade systemic inflammation.The strengths of our study include the broad pre-diagnostic information about the participating prostate cancer patients, a relatively large sample of patients with prostate cancer prostatectomy specimen (n = 142), as well as detailed histopathological and medical records for all the patients. The methodology for TMA-production and in situ hybridization has been used in our lab for several tissues and is well tested [23,24,25,56,57]. Scoring of miR-24-1-5p was completed by two independent observers and showed a moderate inter-observer variability. Earlier studies have focused on murine models and cell lines, while our study uses human prostate cancer tissue, which is in line with future clinical studies. Our study also had some weaknesses. The sample size was not large enough for subgroup analysis, and 42 samples were lost due to technical problems in the ISH-process. The scoring of miRNA-expression was semi-quantitative and thus subject to variability. We only had prostate tissue available and were not able to test the expression of miR-24-1-5p in other samples such as serum or urine.Our study suggests that high expression of miR-24-1-5p is associated with an increased risk of failure after radical prostatectomy, as well as when adjusting for known histopathological risk factors. The results are experimental, based on a relatively small sample size, and should be interpreted with caution. Nevertheless, this could be a steppingstone to further research about the role of miR-24 in prostate cancer and possibly a future tool for better risk stratification.Conceived the study: E.S., E.R., H.S.H. and I.T.; constructed the clinical database: E.S., T.K., M.S., I.T., H.S.H. and E.R.; performed histological examination: E.R.; performed TMA-construction, in situ hybridization and scoring: M.I.P., E.R. and E.S.; performed statistical analyses and drafted the manuscript: E.S., E.R., T.W. and I.T.; critically reviewed the manuscript: E.S., T.W., H.S.H., M.I.P., T.K., M.S., E.G., A.E.E., I.T. and E.R. All authors have read and agreed to the published version of the manuscript.This work was supported by UiT The Arctic University of Norway and the Northern Norway Regional Health Authority grant number SFP1273-16 and SFP1285-16. The Tromsø Study was supported by: the Research Council of Norway; the Norwegian Council on Cardiovascular Disease; the Northern Norway Regional Health Authority; the University of Tromsø; the Norwegian Foundation for Health and Rehabilitation; and the Odd Berg Research Foundation.The PROCA-life study was approved by the Regional Committees for Medical Health Research Ethics (REK Nord), reference no.: 2015/1059, approved date 11 June 2015. The study was conducted according to the guidelines of the Declaration of Helsinki.All participants gave written informed consent upon initial entry to the Tromsø Study, including permission for linkage to other health and medical registers.Dataset available pending permission from the Tromsø Study. Please send request to first author.The authors declare no conflict of interest.TMA blocks were sectioned at 4 µm thickness and mounted on Superfrost Pluss glass slides. During incubations in the instrument, Liquid Coverslip oil was used to protect sections from drying and to ensure proper distribution of reagents.Deparaffinization was performed at 68 °C with EZ Prep solution in three cycles. Target unmasking retrieval was performed at 95 °C with a CC1 buffer to improve the hybridization of the DIG-labeled LNA probes to the patient microRNA sequence. Sections were rinsed with Reaction Buffer between incubations.Target microRNA 24-1-5p: positive control U6snRNA and negative control scramble miRNA probes were diluted in microRNA ISH buffer and Elix RNAse free water to their final concentrations. To achieve optimal hybridization conditions, the probes and tissue microRNA were denaturated for 8 min at 90 °C.Hybridization of the LNA-probes was performed for 60 min. in temperatures adjusted with RNA Tm as a guideline for each probe (Table A1). To ensure specific bindings, stringent washes were completed in two cycles with RiboWash buffer. Antibody Block solution was used for additional blocking against unspecific bindings.For detection of tissue microRNA, anti-DIG-AP Multimer (Alkaline phosphatase (AP)-conjugated anti-DIG) was incubated for 32 min to bind the Digoxygenin-labeled probes. Blue chromogenic visualization of the AP-DIG complex was developed with NBT/BCIP from the ChromoMap Blue detection kit.After Red II counterstain, sections were dehydrated by increasing gradients of ethanol solutions to Xylene and then mounted with Histokitt mounting medium. Ordering details of essential products used in this study are presented in Table A2.Optimized ISH parameters for target probe and controls.Ordering details of products for in situ hybridization.Flow chart of prostate cancer cases. The PROCA-life study (1994–2018).Panel of ISH stained cores. Representative scoring of miR-24-1-5p in tumor epithelium (TE): (A) weak expression, (B) moderate expression, (C) strong expression, (D) U6 positive control staining, and (E) scrambled miRNA negative control staining. The PROCA-life study.Recurrence-free proportion of prostate cancer after prostatectomy, dichotomized into high vs. low expression of miR-24-1-5p in prostate cancer tissue (tumor epithelium and normal epithelium combined (TE+NE)). Low expression was defined as score < 1.57 and high expression as score ≥ 1.57.Distribution of selected characteristics among the prostate cancer patients who received prostatectomy in the PROCA-life Study (1994–2018).* (5 of 52 patients with lymph node dissection). Numbers may vary due to missing information. Values are mean (standard deviation) unless otherwise specified. Abbreviations: PSA, prostate-specific antigen; ISUP, International Society of Urological Pathology. Prostate cancer risk group definitions: low: PSA < 10 µg/L, clinical T-stage (cT-) 1, and ISUP grade group 1. Intermediate: PSA: 10–20 µg/L, cT-stage 2, or ISUP grade group 2–3. High: PSA: > 20–100 µg/L, cT-stage 3, or ISUP grade group 4–5.Distribution and mean score (SD) of miR-24-1-5p expression in prostate cancer tissue by selected characteristics and their subgroups. The PROCA-life study (1994–2018).Numbers may vary due to missing information. Values are mean (standard deviation) unless otherwise specified. Abbreviations: CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical Score.Five-year recurrence free survival (%) for prostate cancer patients after prostatectomy by selected characteristics for all cases and by a subgroup with systolic BP ≥ 130 mmHg. The PROCA-life study (1994–2018).* Log rank test for difference between groups during follow-up until study end. Numbers may vary due to missing information. Regarding miR-24-1-5p: low score was defined as <1.57 and high score ≥ 1.57. Prostate cancer risk group definitions: low: PSA < 10 µg/L, clinical T-stage (cT-) 1, and ISUP grade group 1; intermediate: PSA: 10–20 µg/L, cT-stage 2, or ISUP grade group 2–3; and high: PSA: > 20–100 µg/L, cT-stage 3, or ISUP grade group 4–5. Abbreviations: BP, blood pressure; CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical Score; ISUP, International Society of Urological Pathology; CI, Confidence Interval; TE, tumor epithelium; NE, normal epithelium.Multivariable adjusted * hazard ratio of recurrence of prostate cancer after radical prostatectomy for all cases and for a subgroup of systolic BP ≥130 mmHg. The PROCA-life study (1994–2018).* Adjusted for age, Capra-S group, MiR-24 expression, BMI, kg/m2, cholesterol, and alcohol use. Regarding miR-24-1-5p: low score < 1.57 and high score ≥ 1.57. Abbreviations: Sys, systolic; BP, blood pressure; CAPRA-S, Cancer of the Prostate Risk Assessment Postsurgical Score; CI, confidence interval; TE, tumor epithelium; NE, normal epithelium.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01143.txt

@@ -0,0 +1 @@
+These authors contributed equally as last authors to this work.Non-Small-Cell Lung Cancer (NSCLC) is the primary cause of cancer-related death worldwide. Patients carrying Epidermal Growth Factor Receptor (EGFR) mutations usually benefit from targeted therapy treatment. Nonetheless, primary or acquired resistance mechanisms lead to treatment discontinuation and disease progression. Tumor protein 53 (TP53) mutations are the most common mutations in NSCLC, and several reports highlighted a role for these mutations in influencing prognosis and responsiveness to EGFR targeted therapy. In this review, we discuss the emerging data about the role of TP53 in predicting EGFR mutated NSCLC patients’ prognosis and responsiveness to targeted therapy.Non-Small-Cell Lung Cancer (NSCLC) is the primary cause of cancer-related death worldwide. Oncogene-addicted patients usually benefit from targeted therapy, but primary and acquired resistance mechanisms inevitably occur. Tumor protein 53 (TP53) gene is the most frequently mutated gene in cancer, including NSCLC. TP53 mutations are able to induce carcinogenesis, tumor development and resistance to therapy, influencing patient prognosis and responsiveness to therapy. TP53 mutants present in different forms, suggesting that different gene alterations confer specific acquired protein functions. In recent years, many associations between different TP53 mutations and responses to Epidermal Growth Factor Receptor (EGFR) targeted therapy in NSCLC patients have been found. In this review, we discuss the current landscape concerning the role of TP53 mutants to guide primary and acquired resistance to Tyrosine-Kinase Inhibitors (TKIs) EGFR-directed, investigating the possible mechanisms of TP53 mutants within the cellular compartments. We also discuss the role of the TP53 mutations in predicting the response to targeted therapy with EGFR-TKIs, as a possible biomarker to guide patient stratification for treatment.Lung cancer (LC) is the main cause of cancer-related death worldwide [1]. Non-Small Cell Lung Cancer (NSCLC), the most common LC histology, is a heterogeneous malignancy comprising molecular subtypes for which targeted agents are available in clinical practice [2]. Epidermal Growth Factor Receptor (EGFR) is the most common altered targetable gene in NSCLC, and its mutations (mainly exon 19 deletions and exon 21 L8585R mutations) represent the predictive biomarker for first-, second- and third-generation EGFR-Tyrosine Kinase Inhibitors (TKIs). Nonetheless, resistance to EGFR targeted therapy inevitably occurs, and molecular mechanisms at the basis of primary and acquired resistance to TKIs still have to be elucidated [3].Tumor protein 53 (TP53) gene encodes for p53 protein, a transcription factor recognized as a master regulator of a wide range of cellular processes such as proliferation, differentiation, apoptosis, metabolism and DNA repair [4]. Gene alterations affecting TP53 are the most commons across all cancers, and are involved in cancer onset, development, progression and response to therapies, thus also affecting the patient’s prognosis [4].To date, several TP53 mutations have been highlighted, with relative peculiar protein associated functions; more than 70% of these alterations are represented by missense mutations along the DNA-binding domain (DBD), resulting in different consequences at a cellular, organismal and clinical level [4]. Gene alterations affecting TP53 are proved to be a strong prognostic factor for NSCLC [5], and recent reports indicate a role for these mutations in predicting EGFR-mutated NSCLC patients responsiveness to TKIs. In this review, we focus on the relation between TP53 mutations and EGFR-mutated NSCLC subtype, discussing the achieved results on the role of such mutations in predicting responsiveness to TKIs. We also explore possible cellular mechanisms that TP53 mutants activate to guide the resistance to therapy and discuss the emerging data on the role of these gene mutations for a possible patient stratification for EGFR-mutated patients. Moreover, we discuss the classification systems proposed to date, as it has been demonstrated that different mutations confer different characteristics to the cancer cell.LC is the main cause of cancer-related mortality worldwide (18.4%), while NSCLC accounts for 80–85% of lung cancers [1]. In patients affected by metastatic NSCLC, clinical guidelines recommend the testing of activating mutations, the majority of which are linked with an activation of EGFR that occurs in 10–20% of Caucasian and 50% of Asian patients [1,6].The EGFR gene is located in the short arm of chromosome 7 (7p11.2) [7]. EGFR belongs to the HER/ErbB2 family, a group of receptor tyrosine kinases that include epidermal receptor tyrosine kinases 1 (EGFR, ERBB1), HER2/ERBB2, HER3/ERBB3 and HER4/ERBB4 [8]. These receptors share a similar structure and are composed of three regions: the intracellular, the extracellular, and the transmembrane regions [9]. When the receptor is activated by the ligand, it dimerizes and autophosphorylates the tyrosine residues in the cytoplasmic domain. This step consequently allows the triggering of the intracellular signaling and gene transcription process [9]. The intracellular signaling cascade is mediated through the following pathways: RAS/RAF/MEK/MAPK, PI3K-AKT, JAK/STAT, Src kinase, the Endocytic pathway [10]. Downstream EGFR signaling influences gene expression, apoptosis inhibition, proliferation, angiogenesis, cell motility, and metastasis [6,7,9,11].The EGFR activity can be dysregulated by several activating mutations that occur within the exons from 18 to 21 (encoding the kinase domain) [8]. Exon 19 deletions (ex19Del) of amino acids 747–750 account for 45% of all EGFR mutations, while exon 21 mutations account for 40–45% and are characterized by the substitution of leucine for arginine (L858R) [11]. The remaining 10% are uncommon mutations affecting exons 18 and 20. Among these, the most frequent (4–8%) is the exon 20 insertion, which is also associated with resistance to the three generations of EGFR inhibitors [6]. Ex19Del and L858R are responsible for a constitutional activation of the receptor: ex19Del shortens the activation loop and prevents the rotation of the alpha helix causing a destabilization of the inactive conformation. L858R causes the interaction between the N-lobe and the C-lobe in the inactive conformation. L858R causes steric hindrance and leads to a constitutive active conformation [6].Gefitinib and erlotinib are first-generation reversible EGFR inhibitors. The activity of gefitinib and erlotinib were initially evaluated in unselected NSCLC patients with poor results [12]. Despite these disappointing results, a retrospective analysis of responders to these treatments allowed the researchers to identify EGFR mutations as predictive biomarkers for responsiveness to EGFR TKIs.The IPASS trial was the first randomized phase III trial evaluating the efficacy of gefitinib versus chemotherapy (carboplatin/docetaxel) in 1217 treatment-naïve EGFR-mutated NSCLC patients. This study met its progression-free survival (PFS) primary endpoint and objective response rate (ORR) endpoint [13]. Indeed, the PFS (9.5 months vs. 6.3 months; HR 0.48, p < 0.001) and ORR (71.2% vs. 47.3%) in the gefitinib arm was superior to those of the chemotherapy group. It is of note that this study confirmed that EGFR mutations are the strongest predictive biomarker for response to front-line gefitinib. The phase III NEJ002 study confirmed the superiority in terms of PFS gefitinib compared to carboplatin and paclitaxel (10.8 months vs. 5.4 months; HR 0.3; p < 0.001) [14]. Similarly, in the phase III WJT0G3405 trial, PFS was higher in the gefitinib arm than in the carboplatin/docetaxel arm (9.2 vs. 6.3 months, HR 0.489, p < 0.0001) [15]. Moreover, two phase III trials compared erlotinib to chemotherapy confirming the superiority of this EGFR TKI in terms of PFS [16,17]. As shown in Table 1, a total of six large phase III trials showed a strong benefit of EGFR TKIs versus chemotherapy as a first-line treatment in terms of PFS and ORR in patients with EGFR-mutated NSCLC [13,14,15,16,17,18,19]. However, these studies did not demonstrate a significant benefit in Overall Survival (OS), probably due to the high cross-over rate between the Gefitinib or Erlotinib arm and the chemotherapy arm [20].Based on these results, gefitinib was approved by the FDA (US Food and Drug Administration) in 2015 and erlotinib in 2016 for the treatment of patients with metastatic NSCLC whose tumors have EGFR exon 19 deletions or exon 21 (L858R) substitution mutations.The efficacy of another first-generation EGFR-TKI, namely icotinib, has been evaluated in a phase III CONVINCE trial. In this first-line study, icotinib showed a higher PFS in comparison with the chemotherapy arm (11.2 months vs. 7.9 months; HR 0.61; p = 0.06) [21]. Therefore, based on CONVINCE results, icotinib was approved by the China Food and Drug Administration (CFDA) in June 2011 as a first-line treatment, enriching the EGFR-mutated NSCLC therapeutic armamentarium.The evolution of the EGFR TKIs in the therapeutic landscape entailed the need to define a better first-line strategy treatment in this patient setting. Therefore, the phase III CTONG 0901 trial was conducted to compare the efficacy and safety of gefitinib with that of erlotinib in patients with metastatic NSCLC characterized by EGFR exon 19 or 21 mutations [28]. The results of this comparison demonstrated that the PFS and OS of the erlotinib and gefitinib arms were 13.2 months vs. 11.1 months (HR 0.80; p = 0.108) and 22.4 vs. 20.7 months (HR 0.98; p = 0.902) respectively. Even though this study did not meet its primary endpoint, the subgroup analyses demonstrated that patients with EGFR exon 19 mutations had a significantly higher RR (62.2% vs. 43.5%, p = 0.003) and superior median OS (22.9 vs. 17.8 months, p = 0.022) than those with exon 21 mutations treated with erlotinib or gefitinib.The ICOGEN trial is another randomized, double-blind, phase III trial that was conducted to evaluate the safety and efficacy of icotinib and gefitinib in advanced NSCLC patients previously treated with chemotherapy [29]. In this study, the PFS (7.8 months for icotinib vs. 5.3 months for gefitinib, p = 0.32, and OS 20.9 vs. 20.2; p = 0.76) were similar between the two arms, but the toxicity of the icotinib arm was lower than that of the gefitinib arm (60.5% vs. 70.4%). Thus, compared with the results of these two head-to-head studies, the three first-generation EGFR-TKIs do not seem to be significantly different in terms of the PFS and safety profile.Based on the evidence of a link between EGFR stimulation and increased angiogenesis, first-generation EGFR TKIs were tested in combination with anti-angiogenic compounds. An anti-vascular endothelial growth factor (anti-VEGF) +  erlotinib combination has been investigated in patients with untreated, advanced, EGFR mutated NSCLC in the phase II JO25567 trial, which evaluated the efficacy of erlotinib and bevacizumab compared with erlotinib alone in patients with EGFR mutation-positive NSCLC. The PFS benefit was more consistent for the erlotinib + bevacizumab arm than in the erlotinib arm (median, 16.0 vs. 9.7 months; HR, 0.54; 95% CI, 0.36–0.79; p = 0.0015), leading to EMA approval for this combination [30].A phase III trial evaluated the efficacy of the addition of ramucirumab (anti-VEGF receptor inhibitor) to erlotinib, finding a significant benefit in PFS (19.4 vs. 12.4 months, HR: 0.59, p < 0.0001) [31]. Based on these results, despite the fact that the data of OS are immature, this combination has been approved by the FDA (US Food and Drug Administration) for the first-line treatment of patients with metastatic NSCLC harboring EGFR exon 19 deletions or exon 21 mutations. Interestingly, a biomarker analysis from the RELAY trial highlighted that p53 and EGFR co-mutations were associated with prolonged PFS in the experimental arm, both in exon 19 and 21 mutated patients [32]. The addition of chemotherapy to first-generation TKIs was also investigated. A recent meta-analysis of randomized controlled trials that compared EGFR-TKI monotherapy with the combination of EGFR-TKI and chemotherapy showed that there was a benefit in terms of ORR, PFS and OS in favor of the combination arm, with an acceptable toxicity profile. These data are of interest and the combination of chemotherapy with EGFR TKIs could be a potential first-line treatment in selective patients [22].Afatinib and dacomitinib are second-generation irreversible EGFR TKIs with a similar structure with that of gefitinib or erlotinib but with a side chain that binds covalently to cysteine-797 of EGFR with a subsequent irreversible EGFR inhibition [33,34].Afatinib was evaluated as a first-line treatment in comparison with cisplatin and pemetrexed in the phase III LUX-Lung 3 trial, demonstrating superiority in terms of PFS compared to chemotherapy (11.1 months vs. 6.9 months; HR 0.58; p = 0.001). In this trial, only patients with common EGFR mutations (exon 19 deletions and L858R) were considered, with a reported increase in PFS of 13.6 months for the experimental arm vs. 6.9 months for the chemotherapy arm (HR = 0.47; 95% CI, 0.34–0.65; p = 0.001). Interestingly, the PFS result was superior in patients with tumours harboring an exon 19 deletion with respect to the L858R mutation [35]. On the basis of this clinical trial result, afatinib was approved as a treatment for treatment-naïve patients affected by advanced EGFR-mutated NSCLC. In the LUX-Lung 6 trial, patients were randomized to receive afatinib or gemcitabine/cisplatin chemotherapy. The final results of this study showed a statistically significant benefit in terms of the PFS for the afatinib arm (11.0 vs. 5.6 months; HR 0.28; p < 0.0001) [36]. Despite the PFS benefit for afatinib in these two trials, no difference in terms of survival benefit in the overall population was reported. However, in a prespecific EGFR exon 19 subgroup analysis, afatinib demonstrated a significant improvement in OS compared with the chemotherapy group, in both trials. On the other hand, in the LUX-Lung 7 study, a head-to-head comparison between afatinib and gefitinib, no significant differences in terms of PFS were found (11.0 months for afatinib vs. 10.9 months for gefitinib; HR 0.73; p = 0.017) [24].Another second-generation EGFR inhibitor, dacomitinib, was evaluated in the phase III ARCHER-1050 trial [37]. A total of 452 patients with EGFR-mutant NSCLC were randomized to receive as a first-line treatment dacomitinib 45 mg/day or gefitinib 250 mg/day. The results of this trial demonstrated that the PFS of the dacomitinib was statistically longer compared with the gefitinib arm (14.7 months vs. 9.2 months HR: 0.59; 95% CI: 0.47–0.74). In responders, the duration of response was longer in the dacomitinib arm vs. the gefitinib arm in both EGFR mutation subgroups. However, a dose reduction of dacomitinib was still needed in 66.5% of the trial patients due to a more significant toxicity profile of dacomitinib. Moreover, mOS was significantly longer in the dacomitinib arm than in the gefitinib arm (34.1 months vs. 26.8 months, respectively; HR 0.76; 95% CI, 0.582 to 0.993; p = 0.0438). In the updated OS analysis, a significant improvement in survival was confirmed for the dacomitinib group in patients with exon 21 (L858R) substitution mutations (32.5 months vs. 23.2 months; HR 0.67; 95% CI: 0.47–0.94; p = 0.02); whereas, there was no survival benefit in patients with the exon 19 deletion mutation (36.7 months for dacomitinib vs. 30.8 months for gefitinib, HR 0.85; p = 0.30) [23].One of the most common mechanisms of acquired resistance during the treatment with first- and second-generation EGFR TKIs is represented by the mutation of T790M (EGFR exon 20). Therefore, clinical development was focused on second-line treatments with the subsequent development of third-generation EGFR TKIs. Osimertinib is an oral, third-generation, irreversible EGFR TKI designed to have activity against T790M mutations as well as activity in the central nervous system [27]. The phase III AURA3 trial evaluated the effects of osimertinib or cisplatin/pemetrexed in randomized EGFR-positive patients with T790M resistance [38]. This study demonstrated a significant improvement in PFS, with a median of 10.1 months in the osimertinib arm compared to 4.4 months with chemotherapy (p < 0.001). Furthermore, osimertinib was demonstrated to be superior to chemotherapy in terms of ORR (71% vs. 31%) and with an intracranial ORR of 70%. Based on the results of this study, osimertinib was approved as a second-line treatment of EGFR T790M-mutated NSCLC patients.A sequent phase III FLAURA study assessed the efficacy and safety of osimertinib in patients with previously untreated EGFR mutated advanced NSCLC compared with the standard first-generation EGFR-TKIs (gefitinib or erlotinib) [39]. The results of the FLAURA trial showed that osimertinib resulted in significantly longer PFS (18.9 months vs. 10.2 months; HR 0.46; p < 0.001) with a better safety profile with respect to standard EGFR-TKIs. Despite crossover by 31% of the patients in the control arm, osimertinib confirmed its superiority in terms of OS with a median OS of 38.6 months (95.05% CI, 34.5–41.8) vs. 31.8 months (95.05% CI, 26.6, 36.0) for standard EGFR TKI treatment, respectively (HR 0.799; 95.05% CI, 0.641–0.997; p = 0.0462). Based on these practice-changing results, osimertinib was approved as a first-line treatment for patients with metastatic NSCLC harboring EGFR exon 19 deletions or exon 21 L858R mutations.Approximately 10–15% of EGFR-mutated tumors present uncommon somatic EGFR mutations such as exon 18 nucleotide alterations, exon 19 in frame insertions, exon 20 alterations, and exon 21 mutation L861Q. The knowledge concerning the efficacy of EGFR TKIs in patients affected by NSCLC carrying uncommon EGFR mutations is limited to small clinical investigations or case reports. A post hoc analysis of the LUX-Lung 2, LUX-Lung 3, and LUX-Lung 6 trials demonstrated that patients with L861Q, G719X and S768I responded well to afatinib treatment, whereas afatinib did not show any efficacy in patients with an exon 20 insertion mutation [40,41]. Another retrospective study confirmed the minor efficacy of first-generation EGFR TKIs in patients with G718X and L861Q [42]. Encouraging results come from the clinical activity of amivantamab, a bispecific antibody directed against EGFR and MET tyrosine kinase receptors, and against tumors with EGFR exon 20ins mutations [43]. However, we have to wait for the conclusion of ongoing phase III studies to better define its role in this patient setting. Moreover, osimertinib efficacy was tested in a phase II trial of enrolled patients with uncommon EGFR mutations, demonstrating a 50% objective response rate and 8.2 median PFS [44].Despite the known and confirmed efficacy of EGFR-TKIs in the treatment of EGFR mutated NSCLC, about 5–25% of patients, who receive target therapies, show no benefits from this treatment [45]. The cause of this resistance is the presence of mechanisms that, generally, could be grouped into on-target EGFR-dependent and off-target EGFR-independent. Moreover, primary, or secondary resistance mechanisms could be observed, depending on whether they occurred from the beginning of the treatment with EGFR-TKI or after an initial period of response or stability.On-target mechanisms of resistance occur mostly with the use of first- and second-generation EGFR-TKIs at about 50%, compared to 10–15% for third-generation TKI used as a first-line and 20% as a second-line [3]. One of the most known on-target mutations is T790M, characterized by a threonine substituted with a methionine in the 720 position of exon 20. T790M mostly develops as a resistance mechanism to first- and second-generation EGFR-TKIs used as a first-line treatment, occurring in about 49–63% of patients [46,47,48]. Finding this mutation as a diagnosis of a primary mechanism of resistance is rare. In this case, a possible germinal origin of the mutation should be considered [46]. Other reported rare alterations conferring resistance to first- and second-generation EGFR-TKIs are the missense mutations D761Y, L747S and T854A, while EGFR amplification occurs in 8–10% of cases.On the other hand, some on-target missense mutations conferring resistance to osimertinib have been identified. Exon 20 C797S is able to avoid the binding of osimertinib to EGFR, as well as other third-generation TKIs (e.g., rociletinib, narzatinib, olmutinib) [49,50]. This mutation is the most common tertiary mechanism of resistance and accounts for about 10–26% of cases resistant to osimertinib when used as a second-line treatment; although, when it is used as a first-line therapy, its prevalence is 7% lower, though it remains the second most frequent event behind MET amplification [38,51]. The same effect occurs with mutations affecting L792. On the other hand, it was demonstrated that this resistance mutation to third-generation TKIs is sensible to first-generation gefitinib in vitro [46,50,51]. Another mutation interfering with the binding of osimertinib is the substitution in the 718 residue (L718Q, L718V), placed in the adenosine triphosphate (ATP) binding site. Usually, these mutations confer independent resistance to osimertinib, not coexisting with the C797 mutation [52,53]. Anecdotal and rare mutations interfering with osimertinib activity include C797G and mutations in the C796 residue (C796R, G796S, G796D) in exon 20, near C797 [49,52,54,55,56,57]. In the same protein domain, the G719A mutation has been reported in association with osimertinib resistance. Similarly, other exon 20 mutations (G724S and SV768IL) induce resistance to third-generation TKIs [50,53,56,57]. Finally, the amplification of EGFR and a deletion in exon 19 have been described as additional mechanisms of resistance to osimertinib in the II line setting [58,59]. Lazertinib, a third-generation TKI targeting activating and T790M EGFR mutations, showed important clinical activity in EGFR-mutated NSCLC patients as a second-line treatment after progression to a first- or second-generation TKI [60] and its efficacy, in combination with amivantamab after progression to osimertinib treatment, was demonstrated [61].Off-target resistance mechanisms induced by TKI treatment include alterations affecting pathways that bypass signaling activation, in an EGFR-independent manner. As TP53 mutations involved in the process of histologic transformation to SCLC are a resistance mechanism to EGFR-TKIs, this issue is discussed later.MET amplification is one of the most frequent mechanisms of resistance to all generations of EGFR-TKIs, used as both a first- and second-line of treatment.It was reported in 5–22% of cases treated with first- and second-generation TKIs, mostly in association with an EGFR exon 19 deletion, while it is the most common cause of resistance to osimertinib used in II line [62,63,64]. This data has been observed in the AURA3 trial, where about 19% of the cfDNA samples at the progression showed MET amplification, though the percentage was lower in tumor tissue [53,62,64]. In this setting, MET amplification occurs with or without the loss of T790M, and in 7% of cases could be present with a C797S mutation [63,65].The tyrosine kinase receptor Erb2, encoded by HER2, stimulates the activation of the MAPK and PI3K pathways, mediating resistance to EGFR-TKIs. HER2 amplification has been described in 12% of tumor samples of patients treated with first-generation TKI [66], in 5% of cases treated with osimertinib in II line, and in 2% of cases treated in I line [65,66,67]. This last percentage has been found when the cfDNA was analyzed, and no HER2 amplification was detected in the tumor tissue.KRAS mutation as a mechanism of resistance to EGFR-TKIs is very rare. G12S mutation has been described as resistant to osimertinib in II line; other KRAS mutations, such as G12D, G13D, Q61R and Q61K, have been found to confer resistance to osimertinib as well (reported in less than 1% of cases) [63,65,68,69,70]. In the FLAURA study, mutations of KRAS, like G12D and A146T, were described in the cfDNA of 3% of cases [51]; in the study conducted by Schoenfeld on tumor tissue, KRAS mutation G12A was recorded in one case (4%) [54]. Another mechanism of resistance observed is the BRAF V600F; after treatment with osimertinib in II line, it was observed in 3% of the cfDNA samples [65]. The same frequency was also reported in the cfDNA analyzed after progression with osimertinib used in I line [54]. At progression after osimertinib is used as a second-line treatment, the PI3KCA mutations E454K, E452K, R88Q, N345K, and E418K, have been identified as mechanisms of resistance in 4–11% of cases [62,64,68,70]; in another study, the percentage of these mutations noted on tumor tissue was 17% [71]. Moreover, PI3KCA amplification has been detected, using NGS, in the AURA3 study in 4% of the cases [65]. The main clinical trials investigating new therapeutic strategies for patients with progressive disease after treatment with Osimertinib are resumed in Table 2.The 53 KD protein p53, encoded by the Tumor protein P53 (TP53) gene is a master regulator of cell fate and a powerful antiproliferative transcription factor that controls the epigenetic program and dictates the expression of a plethora of target genes in response to multiple external stresses [72]. In a balanced homeostasis cell state, p53 is maintained at low cytoplasmic levels and kept mostly inactivated by the regulatory action of the E3 ubiquitin-protein ligase, MDM2; various cellular stimuli, including DNA damage and replication induced by oncogenic deregulation, releases MDM2-mediated degradation of p53 and promotes p53 activation by its phosphorylation [73].The best understood function of p53 focuses on its DNA-binding ability to induce cell cycle arrest and promote apoptosis [74]. However, p53 also plays a pivotal role in controlling the overall integrity of the genome, as it is often addressed as the “guardian of the genome”. Upon DNA damage, multiple signaling pathways converge to activate p53, either promoting the repair of the damaged DNA sequence or blocking the DNA replication fork, and in doing so, the propagation of genomic instability and mutations [75]. In addition, p53 can limit chromosomal rearrangements by blocking centromere duplication and telomeric dysfunctions that lead to aberrant mitosis [76]. Mainly for this reason, the absence or inactivation of p53 permits cell survival and facilitates aneuploidy, which is a common step towards further accumulation of oncogenic abnormalities. In addition, p53 suppresses shattering genomic rearrangements such as chromothripsis that typically occur when cells have bypassed replicative senescence [77]. Although the extent of chromothripsis contribution to oncogenesis is still open to debate, this phenomenon is significantly more recurrent when TP53 is deleted or mutated [78]. Another important way by which p53 contributes to maintaining genomic integrity is via the suppression of retrotransposon reactivation and mobilization that can lead to mutagenesis throughout the genome [79]. Specifically, p53 binding to long interspersed nuclear element (LINE) elements promotes LINE epigenetic silencing and might protect the cell from transposon-associated mutagenesis [80].The control of genome integrity by p53 extends to multiple layers that are far beyond its well-known function as a transcription factor [72]. Several genome-wide studies have shown that p53 possesses repressive functions that directly depend on the DNA-binding capacity of p53 [81,82]. ChIP-Seq genome scanning revealed the presence of up to 10,000 possible positions in the genome that p53 potentially binds, which are widespread and do not always represent the binding preferences of p53 as a transcription factor [83]. Furthermore, many of the transcriptionally active promoters bound by p53 do not display direct p53-dependent regulation, suggesting the existence of different dynamics through which p53 might regulate the chromatin status and the overall stability of the genome [81]. For example, p53 can bind with high affinity to regions of chromatin in a closed status [84]. These findings may also raise the hypothesis that it is the nucleosomal structure rather than the DNA sequence affinity that dictates the genomic binding pattern of p53. A comparison between normal and cancer cells suggested that enrichment in CpG islands and hypomethylation of the DNA may drive p53 binding, which likely arise from the overall altered structure of chromatin during oncogenic transformation [85]. In a context of DNA damage, p53 can form a complex with the remodeling and splicing factor 1 (RSF1) forming a complex with the histone acetyltransferase p300 [86].How the interplay between the transcription factor and the chromatin remodeling activity of p53, in a normal cellular context or upon the induction of DNA damage, affects the cell fate and is a question currently awaiting answers; it might have important implications when thinking about cancer therapy. In this concern, p53 inactivation that occurs in cancer may be unique in its ability to both favor genomic instability and sustain survival by downgrading p53 activity as a transcriptional repressor.The TP53 gene has long been recognized as the most frequently mutated gene in human cancer [87]. It is, in fact, well accepted that various mutations in the TP53 gene are the most common genetic lesions found in cancer cells, and mutational dysfunction of the p53 protein is a major contributor to cancer development, progression, metastasis, and resistance to therapy. Further, the presence of mutations that abrogate p53 functionality could also predispose a patient to resistance to cancer therapy [88]. Still, no effective medication that can block the oncogenic derangements derived from p53 inactivation has been approved for use in clinic. The inadequacy in restoring tumor suppressor activity of p53 mutants might also depend on the variety of effects that the different p53 mutations have on the cell [4]. Nevertheless, the precise characterization, in terms of functionality and pathological consequences, of the various p53 mutations is particularly relevant for their use as clinical biomarkers and the optimization of therapeutic options.Most frequently, TP53 mutations in cancer cells occur in one allele, while the other allele has been lost or deleted following major chromosomal rearrangements, leading to a loss of heterozygosity (LOH) that results in the expression of the sole mutated TP53 allele [89]. However, a good fraction of tumors do not present with LOH for TP53, indicating that mutations of the gene might not be necessarily the primary driver of oncogenesis, but might occur at later stages and be just one among many critical pathological events that accumulate during a cancer cell’s life. Typical alterations of TP53 include frameshifts (deletions and insertions), nonsense, silent, and missense mutations that may occur throughout the entire gene sequence [4]. Missense mutations are by far the most common alterations of TP53 (>70%), and normally cluster in the DNA-binding domain (DBD, exons 5–8). Minor hotspot mutations may occur in other coding regions of the gene that are still associated with amino acid residues responsible for the interaction of the protein with DNA. Typically, all of these TP53 mutations have been collectively considered equally for their ability to interfere with the tumor suppressor activity of p53, but there are actually profound differences in the classes of mutations, which may produce distinct outcomes [4].The main function of p53 as a tumor suppressor is linked to its ability to induce cell death or to put the cell into a permanent senescent status. However, it is now quite established that several gain of function (GOF) mutations may happen in TP53, and sustains the notion that cancer cells may actually be addicted to mutated p53 [90]. A depletion of mutated p53 leads to cancer cell death, while ectopic expression of mutant p53 promotes survival via increased genomic instability, angiogenesis, and invasion [91,92,93]. Moreover, recent discoveries unveiled the capacity of p53, in certain contexts, to promote cell survival by also sustaining metabolism and maintaining the balance between glycolysis and oxidative stress, thus, limiting the production of reactive oxygen species [94]. Specifically, chances of p53 levels over time rather than its absolute levels are pivotal in driving cell fate and determining how the cell can respond to perturbations [95]. For example, during treatment with the chemotherapeutic agent Cisplatin, the mRNA stability of p53 target genes, which respond to p53 temporal regulation, is the main determinant in deciding whether the cell will survive to treatment [96]. This notion contributes to the question that multiple levels of p53 regulation may exist, and deciphering the complexity of p53 function relies upon the integration of the tumor suppressive activity of p53 and the understanding that deregulation of some elements of the p53 pathway might also provide the tumour with a survival advantage. Metabolic status, the overall mutational profile, and the epigenetic state of the cell are all determinants of how the tumor suppressive function of intact p53 might be restored during cancer treatment.From a clinical perspective, targeting GOF mutations of p53 may have direct effects on the proliferation and survival of cancer cells that are addicted to p53. However, drugs that target the mutant form of p53, either to block GOF activity or to restore the tumor suppressive activity of p53, should have little interference on the wild-type p53. In particular, restoration of the wild-type function of p53 has been heavily pursued [97]. After almost four decades of studying, p53 is still considered undraggable, especially for the numerous off-target effects that many compounds, initially found as being able to target specifically p53 mutants, actually have. In particular, drugs that promote degradation of mutant p53, have adverse effects on many ubiquitous cellular pathways [98]. Similarly, use of non-selective anti-MDM2 inhibitors that should act in enhancing wild-type p53 activity have been revealed to be problematic due to many adverse side effects in patients [98].This negative trend might, however, soon change, thanks to the availability of novel FDA-approved drugs, already being tested in clinical trials, that are more specific for individual p53 mutations that stratify with patient characteristics [4]. The evolving understanding of the specific functions of the different p53 mutant types might open the door, in the near future, to effective mutant p53 directed therapeutic strategies that will benefit over 50% of cancer patients, particularly those patients with TP53 mutations.As with the majority of cancers, TP53 is the most common mutated gene in NSCLC, also [99,100,101] showing a predominant clonal expression [102]. Data analysis from The Cancer Genome Atlas (TCGA) database highlights that TP53 mutations occur in the exons 4–8 of the gene in 44.8% of cases, confirming that DBD represents the hotspot of the protein. The authors showed that TP53 mutations are able to affect the prognosis of NSCLC patients (OS 27 vs. 19 months, p < 0.001); moreover, different mutations result in different survival times, suggesting that different mutations play different roles at a molecular level [5]. Several studies identified that mutations affecting TP53 are the most influencing prognostic factor, both in early and advanced NSCLC [103,104,105,106]. Moreover, it has been recently reported that activation of TP53 is involved in the EGFR-signaling pathway and in the apoptosis process induced by platinum-based chemotherapy [107]. This evidence, together with the fact that TP53 mutations are more frequent in EGFR-mutated patients, suggest that some of these oncogene-addicted tumors could possess an underlying biology and molecular mechanisms based on two main biomarkers to guide cancer progression [5,108]. On the other hand, in an attempt to find predictive biomarkers for a neo-adjuvant therapy for stage II–III EGFR-mutated NSCLC, exon 4/5 TP53 missense mutations have been found to be a stratification factor for OS and treatment [109]; another study based on the IALT trial case series, found that TP53 mutations play a role in predicting the efficacy of adjuvant platinum-based chemotherapy [110]. In the next paragraph, we discuss the emerging role of TP53 in EGFR-mutated patients, both in terms of prognostic impact and resistance to targeted therapy.The role of TP53 mutations in EGFR-mutated NSCLC has been widely investigated in recent years. One of the first attempts to establish the role of TP53 mutations in this subset of patients was performed by Molina-Vila and colleagues, who explored the prognosis of 125 wild-type (wt) EGFR and 193 (training cohort) and 64 (validation cohort) mutated EGFR NSCLC patients. The authors categorized TP53 mutations as disruptive and non-disruptive ones. Disruptive mutations were identified in stop codons along all the protein structures and non-conservative mutations within the DBD, while non-disruptive mutations were conservative alterations, as well as non-conservative mutations outside the DBD, apart from the stop codons. Their results showed that TP53 non-disruptive mutations are an EGFR- and KRAS-independent prognostic factor (OS 13.3 months vs. 24.6 months, p < 0.001); they also showed that non-disruptive mutations are a prognostic but not a predictive factor in the subgroup of EGFR-mutated patients (median OS 17.8 months vs. 28.4 months in the training cohort, p = 0.004; median OS 18.1 months vs. 37.8 months in the validation cohort, p = 0.006), highlighting a slight trend in the PFS in a multivariate analysis of erlotinib-treated patients (PFS 11.0 vs. 15.0 months, p = 0.14) [111]. Similar results were recently achieved by Aggarwal and colleagues, who highlighted a worse OS for 114 EGFR and TP53 co-mutated patients treated with first-line TKIs with respect to sole EGFR patients (median OS 33.3 months vs. 53.6 months; p = 0.021), with a trend when the data were adjusted for age, smoking status, and performance status [112]. A large cohort study showed a trend for TP53 mutations in predicting a worse OS of EGFR-mutated patients receiving targeted therapy with respect to wt TP53 patients (2.9 years vs. NR, p = 0.06); the trend became more evident when categorizing TP53 mutations as disruptive and non-disruptive (p = 0.055), reaching statistical significance when pooling data from patients with targetable alterations in EGFR, ALK and ROS1 genes (mOS 2.6 years vs. NR, p = 0.009) [113]. In the attempt to find predictive biomarkers for EGFR-mutated NSCLC patients, our group highlighted that mutations in TP53 are able to influence responsiveness to first-line first- and second-generation TKIs in a case series of 136 EGFR-mutated NSCLC patients. In particular, TP53 exon 8 mutations were able to identify a subset of patients with worse DCR with respect to wt exon 8 patients (41.7% vs. 87.3%, p < 0.001) PFS (4.2 vs. 16.8 months, p < 0.001) and OS (7.6 months vs. NR, p = 0.006), with respect to TP53 wt and mutated TP53 in other exon patients [114]. Later, we confirmed our results in an independent case series of EGFR-mutated patients (HR for PFS 3.16, 95% CI 1.59–6.28, p = 0.001) [115]. These results were confirmed in a larger case series of NSCLC patients, where TP53 exon 8 mutations were able to predict the prognosis of EGFR-mutated patients, independently of the received treatment [116]. These results were not confirmed by the abovementioned study, which found that TP53 in exon 8 were not predictors for PFS compared with mutations in other exons (13 months vs. 13.1 months, p = 0.2) [112]. A study by Labbè and colleagues in a cohort of 60 mutant EGFR patients found that TP53 mutations were associated with worse PFS to first-line TKIs only when considering missense mutations (HR 1.91, CI 1.01–3.60, p = 0.04) [117]. These studies underline the importance of a categorization of TP53 mutations; considering that, when taken together, TP53 mutations were not able to reach significant associations with clinical outcome in advanced or early stage NSCLC, even in patients treated with third-generation TKIs [106,111,114,115,117,118,119,120]. The same observation was also highlighted in EGFR-mutated NSCLC patients by Jin et al., who found that TP53 had the higher co-mutation rate with respect to other genic alterations (72.5%), but showed no associations with survival [121]. On the other hand, studies on Asian patients found that TP53 mutations are able to predict PFS of EGFR-mutated patients treated with both first- or second- and third-generation EGFR-TKIs (HR: 2.02; 95% CI: 1.04–3.93, p = 0.038 and HR: 2.23, 95% CI 1.16–4.29, p = 0.017, respectively) [122]. The same results were achieved in a small case series of gefitinib-treated patients, where TP53 mutations affected exclusively short- or intermediate responders (66.7% vs. 0, p = 0.009), and by a study by Yu and colleagues, who found that TP53 mutations in pre-treated patients predicted shorter time-to-progression (HR 1.7, p = 0.006) [123,124]. An interesting study by Roeper and colleagues found that TP53 mutations have a strong negative impact on the clinical outcome of EGFR-mutated patients (ORR, PFS and OS) whether considered as disruptive or non-disruptive; pathogenic or non-pathogenic; or exon 8 or non-exon 8 [125]. Another study found that TP53 mutations are associated with early resistance to EGFR-TKIs, influencing the prognosis of short-responders to TKIs (<6 months, TP53 mutations found in 87.5% of short-responders and in 16.7% of long-responders, p = 0.0002) [126]; these results were later confirmed by another study, which found that TP53 mutations influences the prognosis of short-responders (<3 months, HR = 1.87, 95% CI 1.06–3.29, p = 0.03) and short-survivors (<6 months, HR = 2.73, 95% CI 1.20–6.21, p = 0.017) [127], and by another study, which found that 100% of non-responders to gefitinib were TP53 mutated, with respect to 39% of responders (p < 0.001) [128]. It has recently been demonstrated that TP53 mutations are independently associated with PFS in both first-, second-, and third-generation EGFR-TKIs (HR: 2.02; 95% CI: 1.04–3.93, p = 0.038 and HR: 2.23, 95% CI 1.16–4.29, p = 0.017, respectively) [122]. An interesting study by Tsui and colleagues explored the dynamics of resistance mechanisms to EGFR-TKI treatment through circulating DNA, finding that pretreatment of TP53 mutations predicted a worse OS of patients (HR 0.43, 95% CI 0.2.0.97, p = 0.035), and that TP53 mutation was observed in a patient experiencing progressive disease and without T790M [129]. Results from positive studies demonstrating a role of TP53 mutations in predicting the clinical outcome of EGFR-mutated patients are resumed in Table 3.It has been highlighted that the different methodologies used to detect TP53 mutations have different sensitivities, as next-generation sequence technologies have led to an assessment of a greater rate of mutations in the case series; moreover, massive parallel sequencing has led to the sequencing of the entire TP53 gene, highlighting mutations out of the DNA binding domain.Two meta-analyses were also conducted, to better elucidate the role of TP53 in mutant EGFR patients. The pooled results from 11 studies (1049 EGFR-mutated patients) highlighted that patients with concomitant TP53 mutations have worse PFS (HR 1.76, 95% CI 1.44–2.16, p < 0.001) and OS (HR 1.83, 95% CI 1.47–2.29, p < 0.001). To better estimate the role of such mutations in predicting the response to EGR-TKIs, the authors performed a subgroup analysis of TKI-treated patients. The results remained significant both for patients receiving a TKI as a first-line therapy (HR for PFS 1.69, 95% CI 1.25–2.27, p = 0.001; HR for OS 1.94, 95% CI 1.36–2.76, p < 0.001) and for patients receiving TKIs in an all lines setting, for PFS and OS (HR for PFS 1.99, 95% CI 1.62–2.44, p = 0.001; HR for OS 1.93, 95% CI 1.45–2.58, p < 0.001) [130]. Another meta-analysis based on 2979 EGFR-mutated patients confirmed a worse OS for TP53 mutated patients versus wt TP53 patients (HR 1.73, 95% CI 1.22–2.44, p = 0.002). Interestingly, in a subgroup analysis, patients treated with TKIs had a worse PFS compared to other patients (HR 2.18, 95% CI 1.42–3.36, p < 0.001), even though this data was not confirmed by ORR analysis (RR 1.15, 95% CI 0.92–1.44, p = 0.212) [131]. Even though, in these studies, a categorization of TP53 mutations was not performed, a clear role for TP53 mutations as a negative factor for PFS and OS in EGFR-mutated patients treated with EGFR-TKIs was highlighted. In a study based on TCGA data, Wang et al., found that patients carrying TP53 and EGFR co-mutations had a worse OS with respect to wt/wt patients (38.4 months vs. 51.9 months, p  =  0.023) [132]. Other indications highlight that TP53 mutations are able to influence the response to TKIs depending on the type of EGFR alterations. Considering that patients with EGFR exon 19 deletions usually have a major benefit from first-line EGFR-TKIs compared to other EGFR mutations, it has been reported that TP53 mutations are able to mainly affect the ORR and PFS of this subgroup of patients [114,133,134,135]. On the other hand, a recent study from the BENEFIT trial cohort investigated the TP53 mutations in liquid biopsies of EGFR-mutated patients and found that exon 19 deletions and TP53 co-mutations were predictors of better PFS and OS compared to exon 21 L858R/TP53 co-mutations (HR for PFS 1.53, p = 0.02; HR for OS 0.77, p = 0.37). The same study also demonstrated a shorter PFS and OS for TP53 mutated patients vs. TP53 wt ones (HR for PFS 0.66, 95% CI 0.48–0.89, p = 0.007; HR for PFS 0.54, 95% CI 0.40–0.74, p < 0.001), with TP53 mutations affecting the PFS and OS of patients with EGFR L858R mutations more than those with an exon 19 deletion. Interestingly, the authors also highlighted that patients with TP53 exon 6 and 7 mutations experienced worse PFS and OS, with a role in predicting prognosis for exon 5 TP53 mutations when categorized as disruptive and non-disruptive (OS HR = 2.04, 95% CI 0.99–4.19, p < 0.005) [25]. Recent results based on the CTONG 0901 trial identified that patients with TP53 mutations in exons 4 or 7 experience worse PFS and OS with respect to patients with mutations affecting other exons of the gene or wild-type TP53 (PFS 9.4, 11.0, and 14.5 months, respectively p = 0.009; OS 15.8, 20.0, and 26.1 months, respectively p = 0.004) [136]. The frequencies of TP53 mutations in EGFR-mutated patients are reported in Figure 1.TP53 mutations also play a pivotal role in the histologic transformation to small-cell lung cancer (SCLC), a mechanism of resistance to first- and second-generation EGFR-TKIs known since 2011 [62]. The inactivation of genes like RB1 and TP53 play a crucial role in this transformation, as RB1 and TP53 co-alterations present in patients’ tissues at baseline are more likely to predict histologic transformation, with respect to the presence of only one mutation, and the presence of both mutations is a significant prognostic factor [137,138,139]. In this setting, it seems that TP53 mutation confers genomic instability to cancer cells resulting in a facilitated cell plasticity and phenotype reprogramming. The frequency of this evolution is about 3–5% and, usually, EGFR mutation is maintained [139,140,141].Transformation to SCLC has also been observed as a mechanism of resistance to third-generation EGFR-TKIs used as a second-line treatment, with a frequency between 2 and 15%. This mechanism has been noticed with loss of T790M, but also when the resistance mutation is maintained, suggesting a focal and clonal tumor evolution [69,142]. When osimertinib is used as a first-line of treatment, transformation to SCLC has occurred in 4% of cases, as indicated by Schoenfeld and colleagues [54]. Transformation toward squamous cell carcinoma has been identified after treatment with osimertinib in II or later line and in I line, with a percentage of 9% and 7% respectively [54].Recent results highlight that different TP53 mutations influence, in diverse ways, the response to EGFR-TKIs in cell lines. In particular, some mutations are associated with primary resistance to EGFR-TKIs, while others can induce epithelial-mesenchymal transition (EMT) as an acquired resistance mechanism [141]. Considering that TP53 participates in the regulation of the EGFR pathway [107], and that different mutations induce different mechanisms in different cell line models, it could be interesting to better investigate the cellular functions of the different TP53 mutants. In this regard, a classification based on the disturbance grade of the p53 protein has been proposed [111], and several studies have confirmed that TP53 mutations have different roles at a cellular level, and that some types of mutations are associated with oncogenic GOF [142]. Even though many studies of different case series brought interesting results based on this classification, which mutations and through which cellular mechanisms remain unanswered questions. Moreover, a study by Wei and colleagues highlights that TP53 mutations are generally involved in resistance and primary and metastatic relapse, but different TP53 exon alterations are involved in different mechanisms [143], and there is evidence that TP53 mutations are found more frequently in association with EGFR mutations [144,145,146], suggesting that a subset of EGFR-NSCLC could have a “double-oncogene” addition, for the role that such mutations display.NSCLC is the primary cause of cancer-related deaths. Several pieces of evidence suggest that TP53 mutations are able to be used to identify a subset of patients with worse prognosis and a worse response to EGFR-TKIs; thus, identifying the different mutations associated with different outcomes. Given that TP53 exerts its functions through a wide range of cellular pathways, it would be necessary not only to understand the function of a single mutation, but to consider how this mutation could affect which pathway, though which cellular and molecular mechanisms lie at the base of these processes still has to be elucidated. For this reason, the more that is revealed about the role of these gene mutations in NSCLC, the better we will understand the role in primary or acquired resistance to EGFR-TKIs in this malignancy. What we can assume so far is that, as observed in other tumors, TP53 is a prognostic factor; however, after many reports concerning its effect on the efficacy of EGFR-TKIs, we now need to understand some of the molecular processes that link such mutations and resistance to EGFR-targeted therapy, as this could guide resistance to therapy. The link between TP53 mutations and targeted therapy response should be considered a starting point for new investigations, and further studies are needed to investigate these mechanisms to effectively predict responsiveness and survival; thus, better tailoring targeted therapy for EGFR-NSCLC patients.Ideation: M.C., L.C., G.B. and P.U. Literature search: M.C., K.A., P.C., L.P. and G.B. Paper draft: M.C., K.A. P.C. and L.P. Revisions: M.C., K.A., I.P., P.C., L.P., M.U., A.D., L.C., G.B., P.U. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The data presented in this study are available in the article.The authors declare no conflict of interest.Frequency of TP53 mutations in EGFR-mutated NSCLC patients, extracted by cBioPortal (https://www.cbioportal.org/, accessed on 28 January 2022). The percentage of TP53 codon mutations along all gene exons, and the most frequently mutated codons are reported. TA: transactivation domain; PR: proline rich domain; DBD: DNA binding domain; OD: oligomerization domain; CTD: carboxy-terminal domain.Phase III trials comparing EGFR inhibitors to chemotherapy in first-line treatment.mPFS: median Progression-Free Survival. mOS: median Overall Survival. ORR: Objective Response Rate. NR: Not Reported.Ongoing clinical trials in Osimertinib-resistant EGFR-mutated NSCLC.EGFR: Epidermal Growth Factor Receptor. NSCLC: Non-Small-Cell Lung Cancer. MDT: maximum tolerated dose. OS: Overall Survival. PFS: Progression-Free Survival. RP2D: recommended phase II dose. ORR: Objective response rate.Studies that found that TP53 mutations are prognostic for EGFR-mutated NSCLC patients treated with EGFR-TKIs.EGFR: Epidermal Growth Factor Receptor; TKI: Tyrosine Kinase Inhibitor; OS: Overall Survival; DCR: Disease Control Rate; PFS: Progression-Free Survival; TTP: Time-to-progression.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01144.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Cancer still constitutes one of the main global health challenges. Novel approaches towards understanding the molecular composition of the disease can be employed as adjuvant tools to current oncological applications. Raman spectroscopy has been contemplated and pursued to serve as a noninvasive, real time, in vivo tool which may uncover the molecular basis of cancer and simultaneously offer high specificity, sensitivity, and multiplexing capacity, as well as high spatial and temporal resolution. In this review, the potential impact of Spontaneous Raman spectroscopy in clinical applications related to cancer diagnosis and surgical removal is analyzed. Moreover, the coupling of Raman systems with modern instrumentation and machine learning methods has been explored as a prominent enhancement factor towards a personalized approach promoting objectivity and accuracy in surgical oncology.Accurate in situ diagnosis and optimal surgical removal of a malignancy constitute key elements in reducing cancer-related morbidity and mortality. In surgical oncology, the accurate discrimination between healthy and cancerous tissues is critical for the postoperative care of the patient. Conventional imaging techniques have attempted to serve as adjuvant tools for in situ biopsy and surgery guidance. However, no single imaging modality has been proven sufficient in terms of specificity, sensitivity, multiplexing capacity, spatial and temporal resolution. Moreover, most techniques are unable to provide information regarding the molecular tissue composition. In this review, we highlight the potential of Raman spectroscopy as a spectroscopic technique with high detection sensitivity and spatial resolution for distinguishing healthy from malignant margins in microscopic scale and in real time. A Raman spectrum constitutes an intrinsic “molecular finger-print” of the tissue and any biochemical alteration related to inflammatory or cancerous tissue state is reflected on its Raman spectral fingerprint. Nowadays, advanced Raman systems coupled with modern instrumentation devices and machine learning methods are entering the clinical arena as adjunct tools towards personalized and optimized efficacy in surgical oncology.According to World Health Organization (WHO) [1], in 2020 nearly 10 million cancer deaths have been accounted worldwide while the most common cancer cases pertain breast cancer (2.26 million cases); lung cancer (2.21 million cases); and colon and rectum cancer (1.93 million cases) [2,3,4]. Therefore, the early and accurate diagnosis as well as the precise and adequate surgical removal of a malignancy can lead to the reduction of cancer’s high mortality rates [5,6,7]. Since the differentiation among benign tumors, premalignant, early-stage malignant and healthy tissue is challenging, repeated biopsies are often necessary. Positive predictive values regarding tissue sampling are as low as 22% for prostate cancer diagnosis, 1.4% for breast cancer, 18.5% in lung cancer screenings and 7–23% for melanoma diagnosis [8,9,10,11]. Various conventional imaging techniques have attempted to serve as adjuvant tools for biopsy and surgery guidance. In the field of ionizing radiation, positron emission tomography (PET), computed tomography (CT) and single photon emission computed tomography (SPECT) offer great results, with undisputable drawback relating to the dose deposition to the patient [12,13,14]. Simultaneously, magnetic resonance imaging (MRI), optical coherence tomography (OCT), white light reflectance (WLR), fluorescence, and high frequency ultrasound by exploiting non-ionizing radiation have proved to be valuable diagnostic tools [15,16,17,18]. Nonetheless, currently, no single imaging modality has been proven sufficient in terms of the required standards of specificity, sensitivity, multiplexing capacity, spatial and temporal resolution, and low cost [19,20]. Moreover, most techniques are unable to provide information regarding the molecular tissue composition [21,22]. They just confide on visual changes of the tissue structure and thus present lack of specificity [23,24].Optical vibrational spectroscopic techniques, such as Raman spectroscopy (RS), can depict the gradual changes among malignant and healthy tissue by exploiting the analysis of the characteristic Raman spectrum [24,25]. Raman spectroscopy is a spectroscopic technique offering high detection sensitivity and spatial resolution of a few μm. In general, RS provides information of the short-range molecular vibrational level where the Raman bands are characteristic of the molecular bonding in each chemical group. Eventually, the chemical conformation and the environment of the macromolecular level determine the exact frequencies of the Raman bands. Therefore, Raman Scattering can provide exquisite detail of particular sites of interest and of any biochemical alteration related to the inflammatory, or cancer state of tissue. These tissue related details are reflected on the spectral fingerprint [26] since Raman spectrum constitutes an intrinsic “molecular fingerprint” of the sample [27,28]. This leads to a treasure of information regarding the vibrational modes related to specific chemical configurations present in tissues, correlated with proteins, lipids, glucose consumption, DNA, RNA, and other biomolecules [29]. Consequently, the entirety of a Raman spectrum can provide the analytic guideline of biological sample’s structure, identity, and composition as well as the depiction of macromolecules interactions and composition [27]. Due to that, RS offers high molecular specificity into the characterization of biological tissues ex vivo and in vitro and constitutes an excellent non-invasive detection method of the molecular differences among tumor and healthy tissue [30,31,32]. Moreover, RS does not require any reagents, labelling or other preparation of the tissue while the use of optical fibers allows the assessment of several anatomical locations in vivo [21,31,33,34].Vigorous attempts have been made during the last decade towards the clinical implementation of Raman spectroscopy in the hope of addressing the same fundamental issue: the inadequacy of pre- and intraoperative methods with satisfactory and clinically relevant specificity and sensitivity. The most recent studies are aiming into the quadruple of: (1) pre-malignant lesions detection, (2) detection of cancer in less advanced stages, (3) the reduction of unnecessary biopsies, and (4) guided surgery for the entire removal of a malignancy with adequate tumor resection margins [35,36,37,38]. According to the literature, in vivo and ex vivo trials which are aiming towards the detection of malignant tissue have accomplished specificities varying between 45–100% and sensitivities varying between 77–100% [39,40,41,42]. Studies with the goal of pre-malignant lesions detection observed sensitivities ranging among 70–93.5% and specificities ranging between 63–97.8% [43]. Even though the numbers presented may not touch perfection, they still constitute a strong argument towards the capability of Raman spectroscopy to enhance current clinical practice.In this review, we provide an overview of the most prominent Raman spectroscopy applications in biological and clinical research. We highlight the perspective of advanced Raman systems incorporation in clinical praxis as an adjunct tool towards early diagnosis and oncologic surgery guidance. The combination of Raman spectroscopy with modern instrumentation devices, novel techniques, and machine learning methods is presented. This coupling will contribute to overcoming current limitations which have prevented the broad clinical application of Raman spectroscopy so far and will establish RS’s potential to be used as a personalized decision-making tool.In general, configurations of nearby chemical bonds are characterized by typical vibrational energies. When photons, emitted by a laser light source, are inelastically scattered by these characteristic molecular oscillations, a Raman scattering event takes place. The detection and analysis of the scattered photons offers a spectrum comparted of the so called characteristic Raman peaks. Each individual peak is indicative of a particular vibrational mode related to distinct chemical configurations [22,25]. Various Raman techniques have been developed to cover the distinct requirements of each biomedical sample such as: Spontaneous Raman Spectroscopy (SRS) [27,44,45], Resonance Raman Spectroscopy (RRS)[45], Surface-Enhanced Raman Scattering (SERS) [46,47,48,49,50], and Coherent anti-Stokes Raman scattering (CARS) [51,52,53,54] etc. Nevertheless, despite the advancement of these techniques, they still present complexities during experimentation and analysis and thus cannot be yet applied as a simple surgery tool. This review will be concentrated on conventional Spontaneous RS. Prominent tissue Raman peaks are observed in the fingerprint of 800–1800 cm−1 and the high frequencies 2800–3200 cm−1, spectral regions. A characteristic example of Raman spectra could be this from colorectal tissues by Bergholt et al. Τhey performed discrimination between normal, hyperplastic, adenoma, and adenocarcinoma using near-infrared Raman spectroscopy [55]. Figure 1 shows the mean of in vivo Raman spectra of normal (n = 1464), hyperplastic polyps (n = 118), adenoma (n = 184), and adenocarcinoma (n = 103) acquired from 121 lesions of 50 patients during colorectal endoscopy. The strongest Raman bands are marked upon the spectra and are related to specific vibrations in cellular or extracellular components: 853 cm−1 (ν(C–C) proteins), 1004 cm−1 (νs(C–C) ring breathing of phenylalanine), 1078 cm−1 (ν(C–C) of lipids), 1265 cm−1 (amide III ν(C–N) and δ(N–H) of proteins), 1302 cm−1 (CH2 twisting and wagging of lipids), 1445 cm−1 (δ(CH2) deformation of proteins and lipids), 1618 cm−1 (ν(C=C) of porphyrins), 1655 cm−1 (amide I ν(C=O) of proteins), 2850 cm−1 and 2885 cm−1 (symmetric and asymmetric CH2 stretching of lipids), 2940 cm−1 (CH3 stretching of proteins), 3009 cm−1 (asymmetric = CH stretching of lipids). Bands above 3200 cm−1 are OH stretching modes of water.The analysis of the vast amount of Raman data is a critical barrier which needs to be overcome in order to enable the facilitation of Raman spectroscopy in the clinical routine [56]. However, the evolution of artificial intelligence (AI) provided a boost in real-time Raman data processing. The combination of AI tools with Raman spectroscopy can efficiently lead to adequate discrimination of cancerous tissues [57]. Machine learning (ML) and Deep learning (DL) constitute branches of the broader division of Artificial Intelligence (AI) [58]. Their advanced innovation deservedly classifies them as an excellent candidate for medical applications especially for those dependent on complex, highly versatile genomic procedures such as cancer diagnosis and detection [59,60,61]. ML or DL could constitute valuable tools in physics applications in medicine such as Raman spectroscopy, where the detection and analysis of various spectral fingerprints is vital [62,63,64]. Figure 2 is a schematic representation of the workflow of the combination of Raman spectroscopy with machine learning models for tissue discrimination and classification using a multilayer perceptron (MLP).In more detail, features extracted from Raman spectra (e.g., representative intensities at certain wavenumbers) and/or from spectral images of biostructures (e.g., pixel intensity patterns) are used as inputs to an MLP. The hidden layers of an MLP will introduce a series of linear and non-linear calculations that will lead to a single output neuron. Each output corresponds to normal or malignant class towards the discrimination between healthy and cancerous tissues. Several machine learning models such as Support Vector Machine (SVM) [65], boosted tress [66], convolutional neural networks (CNNs) [67], and artificial neural networks (ANNs) [68] have been exploited for cancer detection for almost 20 years [69]. Xiaozhou Li et al. focused on the expediency of Raman spectroscopy for colon cancer diagnosis by serum analysis. They observed statistically important spectroscopic differences among cancerous and normal cells for six Raman peaks at 750, 1083, 1165, 1321, 1629, and 1779 cm−1, which indicate nucleic acids, amino acids, and chromophores respectively. The intensity of the peaks in the cancerous cells either increases or decreases reflecting the induced chemical modifications. For example, the decrease of the 1165 cm−1 peak is related to low levels of anti-oxidant β-carotene in the cancerous cells. They used principal component analysis (PCA) and k nearest neighbor analysis (KNN). They concluded that a number of the PC loading peaks are identified as colon tissue peaks which eventually proved the correlation among the original Raman spectra and the PC loading spectra. Specificity calculated by KNN analysis reached 92.6% and a diagnostic accuracy of 91% [70]. The diagnostic models built with the identified Raman bands provided diagnostic accuracy of 93.2% into identifying colorectal cancer.Ragini Kothari et al. investigated rapid, quantitative, probabilistic breast tumor assessment with real time error analysis. They observed that often the spectral shifts that were denoted as malignant would constitute false positives due to lack of lipid signals [71]. Stochastic neural networks (NNs) were exploited to estimate the Bayesian probability of a Raman spectrum containing characteristic peaks of cancer using data from the entire spectral bandwidth (600–3000 cm−1), the fingerprint region (600–1800 cm−1), and the high wavenumber region (2800–3000 cm−1) [68]. Qingbo Li et al. suggested an entropy weighted local-hyperplane k-nearest-neighbor (EWHK) algorithm to determine the Raman spectra in breast cancer by enhancing the classification accuracy [72]. This method led to a positive prediction rate of 95.99%, a negative prediction rate of 83.69%, specificity of 87.77%, accuracy of 92.33%, and sensitivity of 93.81% [72].Shaoxin Li et al. used near-infrared Raman spectroscopy and feature selection approaches to detect colorectal cancer tissues. Significant differences were identified between normal and cancerous cells by using ant colony optimization (ACO) and support vector machine (SVM) algorithms for five Raman bands related to proteins, nucleic acids, and lipids of tissues in the areas of 815–830, 935–945, 1131–1141, 1447–1457, and 1665–1675 cm−1. For example, the 1323 cm−1 band, which is assigned to nucleic acids (CH3CH2 twisting mode), increases in cancer tissues compared to normal ones, reflecting the higher content of nucleic acid in tumor cells. A diagnostic accuracy of 93.2%, a sensitivity of 92.3%, and a specificity of 94.2% were achieved [73].Non-linear NNs have been used to predict the Bayesian probability of breast cancer. Nine spectra regions, six in the fingerprint region (600–1800 cm−1) and three in the high wavenumber region (2800–3200 cm−1), were identified comparing DNA/RNA, protein, carbohydrate, and lipid cellular components of healthy and cancerous cells [71]. Deep convolution neural networks have been applied to fiber optic Raman spectroscopy systems providing a novel classification method for tongue squamous cell discrimination [74] According to the results, high sensitivity (99.31%) and specificity (94.44%) were achieved. According to the literature, Raman spectroscopy-based biopsy guidance presents overall specificities and sensitivities between 66–100% and 73–100% respectively [21,24]. The use of this technique promises a drastic increase in the accuracy of cancer diagnosis and an important reduction in the number of false positive biopsies [31,33,34]. The detector technology improvement, the in vivo fiber-optic probe design and the use of artificial intelligence algorithms as well as the collection of large independent comprehensive datasets obtained in the actual clinical workflow enable the facilitation of Raman-based systems into the routine clinical settings [23].Fiber-optic probes have enabled the access of Raman spectroscopy in in vivo diagnostic techniques [55]. The ability of fiber probes to be inserted endoscopically, especially in hollow and solid organs, such as the upper gastrointestinal tract, the colorectal, and cervical cancers, or the oral cavity, the bladder, and the lung, enables in vivo measurements and in vivo assessment [55,75,76,77]. Advanced fiber probes such as probes with plasmonic nanostructures on their distal end surface can provide enhancement of the surface Raman scattering signal [78]. Moreover, fiber probes can overcome the limited penetration depth of laser radiation in tissues due to the high diffusion and scattering of photons. Figure 3 shows a portable Raman imaging system based on SERS fiber-optics probes capable of conducting white light endoscopy [79].Moreover, novel Raman techniques combined with advanced fiber probes can offer a boost to Raman Spectroscopy’s application in clinical praxis. For example, Micro-scale spatially offset Raman spectroscopy with an optical fiber probe (micro-SORS) can collect photons from deeper layers by offsetting the position of the laser excitation beam [80] and by reaching a penetration depth up to 5 cm [81]. Recently, Zhang et al. combined micro-SORS with Surface-enhanced Raman spectroscopy (SERS) applied on a tissue phantom of agarose gel and biological tissue of porcine muscle [82]. According to their results, the penetration depth could be improved over 4 cm in agarose gel and 5 mm in porcine tissue compared to the 2 cm depth of agarose gel and the 3 mm depth in porcine muscle received by SERS measurements.Stevens et al. and Wang et al. investigated epithelial tissue associated with dysplasia and developed a Raman probe coupled with a ball lens that could enhance in vivo Raman measurements from gastric premalignant epithelial tissue during endoscopy [83,84]. Due to the use of a ball lens, they managed to decrease the collection depth at 300 nm, which is the relevant depth for the analysis of gastric epithelium [83,84]. Moreover, they exploited a multimodal image-guided Raman technique to achieve real time in vivo cancer detection. Bergholt et al. used this high wavenumber system in combination with a foot pedal control switch and auditory feedback to the gastroenterologist during colonoscopy diagnosis [85]. Another team, Agenant et al., developed a novel Raman probe that could take measurements at the depth of 0–200 μm (average urothelium depth), the adequate level for superficial tissue sampling, in order to improve in vivo diagnosis of urothelial carcinoma [86]. This novel probe was comparted of seven collection fibers, one excitation fiber and two component front lens [86]. Figure 4 shows the different geometries of fiber-optics probes used in clinical applications such as endoscopic probes without any focusing optics, confocal endoscopic probes, and fiber probes for side-viewing [87]. Challenges of fiber probe’s use pertain to the intense resemblance between the excitation laser light and the collection of the scattered light by the different tissue’s anatomical regions [88]. Some of the problems arise due to the background Raman and fluorescence signals created by the fiber’s materials and due to the intrinsic fluorescence signal (autofluorescence) of the tissue [54,89]. The separation between the collection and the excitation pathways is still a valid issue for Raman tissue measurements. The background Raman and fluorescence signals created inside the fiber require the separation between the collection and excitation pathway [27]. This generates a challenge regarding the minimum size of such devices. Nijssen et al. attempted to overcome this difficulty by detecting the high wavenumber region from 2500 to 3800 cm−1 (near-infrared region) of the Raman spectrum [47]. That way, the same silica-based fiber optic probe could both guide laser light to the tissue and simultaneously collect scattered light. At the same time, low overlap was achieved with the generated parasitic signals (Rayleigh scattering and Raman from the probe) [47] offering that way a perspective towards the miniaturization of such systems.Moreover, the development of Raman instrumentation regarding in vivo and ex vivo applications is mainly focusing on the overcoming of issues such as: the speed of measurement, the instrumentation cost, and the background interference due to the different types of tissue. Advanced focal-plane detectors, volume-phase holographic gratings, stabilized diode lasers, and imaging polychromators are building a new perspective towards robust Raman instrumentation [90,91] which achieves high quantum efficiency, simultaneous spectral integration from the high spectral and lateral range and low background noise [90,91]. Therefore, the traditional limitations of low sensitivity and poor detection capability that Raman spectroscopy systems used to present are now dropping drastically [92]. New innovative techniques allow infrared and near infrared detection while cutting edge technologies promise system architectures with single photon detection capabilities and hybrid imaging technologies [91,93,94]. The development of an in vertical-external-cavity surface-emitting semiconducting laser presents a large gain area and transverse mode control of the extended cavity, and hence accomplishes a combination of high continuous wave output power and a near diffraction limited beam [93]. Furthermore, semiconductor lasers present the advantages of easy array fabrications and low cost of production [95]. According to current practice, the primary treatment for solid tumors is surgical removal [96,97,98,99]. Adequate surgical margins, vital for disease control, are selected for the resection of the entirety of the cancerous tissue. Of vital importance is the preservation of all healthy structures, in cases where the anatomical regions allow it. However, the surgical resection techniques that are currently used are based on subjective methods, such as visual inspection or palpation to verify the exact margins between malignant and normal tissue [96,97,98,99]. This may lead to partial removal of the malignancies and consequently to the occurrence of residual tumors, strongly correlated with poor survival rates [96,97,98,99]. In addition to that, additional surgeries, or adjuvant therapies such as radiation therapy or chemotherapy may be required. Studies indicate that the five-year survival rate decreases drastically when a solid tumor is not dissected to its entirety [23,96,97,98,99]. Portable Raman systems have been implemented into the clinical environment of oncological surgeries presenting excellent assets such as the ability to offer representative sampling towards correct pathological diagnosis and accurate assistance in the definition of resection margins during surgery. As can be depicted in Figure 5, the objectivity of Raman spectroscopy as an imaging technique collaborated with the data analysis and classification capabilities of Machine Learning techniques could constitute a valuable intraoperative guidance tool.An intraoperative Raman system that directly measures brain tissue in patients has proven to distinguish dense and low-density cancer infiltration from benign brain tissue with a sensitivity of 93% and a specificity of 91% [100,101]. More precisely, the experimental setup was pertaining to a hand-held optic Raman probe and a 785 nm NIR Laser [101]. The research team exploited the boosted trees supervised machine learning algorithm to process their data and eventually differentiate the spectrum among cancerous and healthy brain tissue [101]. In another study, a real-time Raman intraoperative system was used during breast cancer surgery for the assessment of freshly resected specimens [102]. A total of 220 Raman spectra were collected with the aid of an 830-nm-diode laser focused on a Raman optical fiber probe [102]. This study has demonstrated that Raman spectroscopy could discriminate cancerous tissue from normal breast tissue with a sensitivity of 83% and a specificity of 93% [102].A handheld contact Raman spectroscopy probe was used for real-time identification of brain cancer during surgery. Jermyn et al. obtained very fast and high-quality pure Raman signals from 0.5 mm tissue areas with sampling depth up to 1 mm during the tumor resection [66] by using micrometer-scale filters that were placed directly at the tip of the optical fibers [66]. A portable clinical fiber probe system in combination with a classification AI algorithm with the ability to differentiate healthy breast tissue from cancerous tissue was utilized by Barman et al. as a guidance tool for mastectomy procedures. The recorded specificity was 100% with sensitivity of 62.5% [33]. The differentiation among normal, breast cancer, fibroadenoma, and fibrocystic change was achieved with accuracy of 82.2% [33]. In order to reduce the time measurement of whole tissue sections in skin cancer, Kong et al. developed the approach of using auto-fluorescence images at excitation wavelengths of 377 nm and 292 nm in combination with Raman spectroscopy [42]. Since these wavelengths are the corresponding excited peaks of tryptophan and collagen, they managed to differentiate normal dermis (characterized by high collagen expression) to cancerous segments [42]. This method recorded measurements with specificity of 94% and sensitivity of 95% [42]. Short et al. conducted a study using Raman spectroscopy on ex vivo colon tissue from 18 patients, measuring both the fingerprint and high-wavenumber spectral regions [77]. The results indicated that, using the high-wavenumber region, the non-malignant and the malignant groups could be classified correctly with a specificity of 89% [77]. The authors referred that the high-wavenumber region could be used in vivo to improve the identification of neoplastic lesions. In the domain of colorectal cancer, Bergholt et al. using an endoscopic multi-fiber Raman probe measured both the fingerprint and high-wavenumber spectral regions of 50 patients in vivo [55]. The team attempted to differentiate Adenomatous polyps from hyperplastic polyps with a specificity of 83% and a sensitivity of 91% [55]. Table 1 presents an overview of in vivo Raman measurements for clinical applications that have been attempted for a variety of cancer types.The main advantages of RS, such as (a) its non-invasive character and compatibility with tissue physiology due to the weak water signal, (b) its suitability for in vivo fiber-optic applications on versatile cancers, and (c) its high specificity with simultaneous chemical analysis of the malignant tissues have been thoroughly described in the previous sections. There are, however, important limitations of the technique which hinder its establishment in the clinical setting. The most prominent constraints are the following: (a) the weak Raman signals which require long acquisition times, (b) the strong autofluorescence background which affects the quality of the acquired spectra hiding the weak Raman features and complicating the analysis, (c) the potential damage of the tissues by laser heating which is a rather complicated effect depending on the laser excitation wavelength and power, as well as the light absorption coefficient of the tissues, and (d) the subtle differences in the spectra which require sophisticated analysis [147]. Nevertheless, technical advancements in current generation Raman spectrometers and integration of machine learning techniques for big data analysis and cancer classification gradually tilt the balance towards the application of RS as a rapid diagnostic tool in clinical praxis. It is worthwhile to mention that most of the portable Raman spectrometer manufacturers make their own cooperative research in the field and include such applications in their technical notes. Despite the above limitations, Raman Spectroscopy constitutes a very promising technique for in-situ cancer diagnosis. Since the achievement of adequate surgical margins is vital for disease control and survival, an intraoperative guidance tool such as Raman probes will significantly limit the subjective methods which surgical resection techniques are currently based on (such as palpation and visual inspection). However, the effort to reliably assess resection margins in surgical oncology also suffers from constrains inherent in oncological surgery, such as non-standardized practices and the impact of epithelial or other dysplacias at the margins, as well as differences in reporting the status of the surgical margin. With regard to RS applications on excised tissues, the method employed for the retrieval of sections from resection margins or bioptic samples and the issue of postresection tissue shrinkage introduce extra variability. Recent creation of multidisciplinary networks like ClirSpec, Raman4Clinics (EU COST Action BM1401) and EPIC are significantly narrowing the distance among scientists and clinicians [27]. These networks are actively pursuing the standardization of measurements and of the preparation of the samples, the creation of data analysis protocols, and the settlement of a basis for transferability.The compelling recent developments of Raman instrumentation, including the new technologies and the reduction of the cost of lasers, holographic gratings, detectors, and Raman probes as well as the entrance of machine learning models in the analysis of the data have contributed towards the overcoming of some of the method’s deficiencies. Thus, they have inspired numerous research groups towards the elaboration of Raman based techniques with biomedical orientation. Among them, cancer prognosis and diagnosis is an excellent candidate, promising low invasive, in vivo and real time detection, and accurate molecular characterization. RS can achieve specificities of 45–100% and sensitivities of 77–100% in cancer diagnosis and malignancies detections. The objectiveness of RS and the ability to provide biochemical information in combination with the advancement of the Raman probes that can be integrated in endoscopes and provide spectroscopic images of the tissues may be the solution to the fundamental problem of the deficiencies of pre- and intraoperative methods with adequate and clinically relevant specificity and sensitivity. The prospect of adaptation of Raman spectroscopy into the clinical environment could finally provide surgeons with the assurance of the intraoperatively adequate resection margins while it could upgrade patient’s surgical outcome and simultaneously minimize the adjuvant therapies needed.Conceptualization, E.P.E. and E.S.; writing—original draft preparation M.A.K., E.S., D.K., and M.K.; writing—review and editing, M.A.K., E.S., A.G.K., Y.S.R., N.A., N.D., I.S. and E.P.E. All authors have read and agreed to the published version of the manuscript.This research was funded by the European Regional Development Fund of the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH–CREATE–INNOVATE (project code: T1EDK-01223).This research has been co-financed by the European Regional Development Fund of the European Union and Greek national funds through the Operational Program Competitiveness, Entrepreneurship and Innovation, under the call RESEARCH–CREATE–INNOVATE (project code: T1EDK-01223). Title: Development of advanced portable biophotonic system for the personalized spectroscopic discrimination of cancer margins/tissues, Acronym: BIOPHASMA). The authors would like to thank M. Labropoulou and A. Tsaroucha A. for their constructive comments on the manuscript.There is no conflict of interest.Mean ± 1 standard deviation (SD) values of in vivo fingerprint (FP) spectra (800–1800 cm−1) and high-wavenumber (HW) Raman spectra (2800–3600 cm−1) of normal (n = 1464), hyperplastic polyps (n = 118), adenoma (n = 184), and adenocarcinoma (n = 103) acquired from 121 lesions of 50 patients during colorectal endoscopy. The spectra have been normalized to the integrated area in the FP and HW ranges for comparison purpose. Reused with permission from [55]. Copyright 2015 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim, Germany.Depicts the basic structure of Machine Learning workflow applied on a Raman Dataset.Schematic representation of a Raman system with SERS fiber-optic probe based Raman system which can perform white light endoscopy. (a) The design allows the Raman imaging system to get adapted on a clinical endoscope and scan the lumen as the endoscope is being retracted in the GI tract. (b) An expanded schematic illustration of the distal end of the device. The collimated beam can be swept by a brushless DC motor and its focus can be adjusted by a system of plano-convex and plano-concave lenses [79], https://doi.org/10.1371/journal.pone.0123185, access on 10 January 2022 ).Different geometries of fiber-probes used in clinical applications: (a) non-superficial endoscopic probe with one excitation fiber in the center and seven collection fibers arranged around the emitter (b) confocal endoscopic fiber probe with a ball lens (c) fiber probes with mirror (or prisms) [87]. https://doi.org/10.1117/1.JBO.23.7.071210, access on 10 January 2022. PMID: 29956506. Excitation and collection filters are also depicted.Effectiveness of oncologic surgery depends on precisely distinguishing healthy from malignant tissue during the operation. This flow diagram shows the steps of RS- based diagnosis from the patiant examination (a) via the multicomponent instrumentation (laser excitation–Raman probe-scattered light dispersion and detection) (b) in order to acquire the Raman spectra (schematic, not real data) in (c) and towards their analysis and classification via mashine learning techniques (d). A simple multi-layer perceptron neural network architecture is presented. In fact, the input layer is a data matrix with intensity values from different observations at various Raman frequencies. This combined methodology potentially has the ability to accurately differentiate benign from malignant tissue in real time and eventually improve the surgical outcome.Clinical Raman applications for diagnosis and surgery guidance.*s: sensitivity, *sp: specificity, *ppv: positive predictive value, *atd: accuracy of tissue differentiation, *fpsl: false positive suspicious lesions, *ca: classification accuracy, *a: accuracy, *ss: surgical success, *m.r: miss rates, RS: Raman spectroscopy, R.p.: Raman probe.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01145.txt

@@ -0,0 +1 @@
+With nonspecific activation of the immune system, immune checkpoint inhibitors (ICIs) can lead to off-target immune-related adverse events (irAEs) to every organ system. Immune-related cardiotoxicity is rare but often fatal. Large population-based studies examining different ICI-associated cardiotoxicity across cancer types and agents are limited. Using data from a large network of health care organizations, this study aims to: (1) provide an estimate of the incidence of ICI-associated cardiotoxicity, (2) to determine patient and clinical characteristics associated with the risk of developing ICI-associated cardiotoxicity, and (3) to assess the overall survival of patients experiencing ICI-associated cardiotoxicity compared to patients who did not develop cardiotoxicity after ICI use.Large population-based studies examining differences in ICI-associated cardiotoxicity across cancer types and agents are limited. Data of 5518 cancer patients who received at least one cycle of ICIs were extracted from a large network of health care organizations. ICI treatment groups were classified by the first ICI agent(s) (ipilimumab, nivolumab, pembrolizumab, cemiplimab, avelumab, atezolizumab, or durvalumab) or its class (PD-1 inhibitors, PD-L1 inhibitors, CTLA4-inhibitors, or their combination (ipilimumab + nivolumab)). Time to first cardiac adverse event (CAE) (arrhythmia, acute myocardial infarction, myocarditis, cardiomyopathy, or pericarditis) developed within one year after ICI initiation was analyzed using a competing-risks regression model adjusting for ICI treatment groups, patient demographic and clinical characteristics, and cancer sites. By month 12, 12.5% developed cardiotoxicity. The most common cardiotoxicity was arrhythmia (9.3%) and 2.1% developed myocarditis. After adjusting for patient characteristics and cancer sites, patients who initiated on monotherapy with ipilimumab (adjusted Hazard Ratio (aHR): 2.00; 95% CI: 1.49–2.70; p < 0.001) or pembrolizumab (aHR: 1.21; 95% CI: 1.01–1.46; p = 0.040) had a higher risk of developing CAEs within one year compared to nivolumab monotherapy. Ipilimumab and pembrolizumab use may increase the risk of cardiotoxicity compared to other agents. Avelumab also estimated a highly elevated risk (aHR: 1.92; 95% CI: 0.85–4.34; p = 0.117) compared to nivolumab and other PD-L1 agents, although the estimate did not reach statistical significance, warranting future studies.Immune checkpoint inhibitors (ICIs) are monoclonal antibodies that activate T cells and initiate an adaptive immune response, thereby allowing the immune system to recognize abnormal cancerous cells [1]. Currently, there are seven FDA-approved ICIs targeting three pathways: cytotoxic T-lymphocyte associated-antigen-4 (CTLA-4) (ipilimumab), programmed death receptor-1 (PD-1) (pembrolizumab, nivolumab, and cemiplimab), and programmed death-ligand-1 (PD-L1) (atezolizumab, avelumab, and durvalumab) [2]. Durable tumor responses and improvement in overall survival have been shown in numerous randomized clinical trials in patients treated with ICIs [1]. It was estimated that 43.6% of cancer patients in the US were eligible for ICI therapy in 2018 [3]. Since then, the FDA has approved more indications for ICIs [2] and the number of eligible patients is likely to be even higher. However, not all patients respond to ICIs and the overall response rate (ORR) varies by tumor type and agent [3], ranging from 10.9% for single-agent ipilimumab in previous treated melanoma [4] to 69% for pembrolizumab in relapsed/refractory classic Hodgkin’s lymphoma [5], with a subset of patients developing resistance over time [1]. To increase response and combat resistance, combinations of ICIs with each other or with treatments such as chemotherapy, radiation, and targeted therapy are increasingly being used which increases the complexity of toxicity [6]. The use of ICIs is rapidly expanding. According to the 2018 estimates by Allied Market Research, globally, the ICI market was valued at $29.8 billion in 2020, and is projected to reach $140.9 billion by 2030, at a compounded annual growth rate of 16.8% between these years [7].With nonspecific activation of the immune system, ICIs can lead to off-target immune-related adverse events (irAEs) in every organ system [8]. It is estimated that up to 90% of patients experienced any clinically detectable irAEs and as many as 45% of patients experienced severe irAEs (grades 3–4), although the estimates vary by agents and are higher among combination therapy [9]. Immune-related cardiotoxicity is rare but often fatal [10]. The most commonly reported cardiac irAE from ICIs is myocarditis, an inflammatory disease of the heart muscle cells [10]. Existing studies have reported a wide range of incidence of ICI-associated myocarditis, ranging from 0.06% for single-agent nivolumab and 0.27% for a nivolumab and ipilimumab combination in the Bristol Myers Squibb corporate safety database [11] to 1.14% across multiple ICIs and combinations in a single-center study [12]. Both studies were conducted when ICI-associated cardiotoxicity began to be recognized [11,12]. Pharmacovigilance studies have since raised its awareness [13,14], leading to more reported cases in recent years [15]. However, diagnosis of myocarditis is challenging due to its heterogeneous clinical manifestations, which range from no symptoms with an abnormal biomarker, to nonspecific symptoms such as fatigue, to fulminant presentation with hemodynamic compromise [1,16]. Moreover, to confirm diagnosis, the gold standard is endomyocardial biopsy, which is an invasive procedure and not often performed in clinical practice [16]. Thus, the incidence of ICI-associated myocarditis is likely to be underestimated. Unlike general myocarditis, ICI-associated myocarditis is highly fatal, with a reported mortality rate of 40–50% [13,17]. In addition to myocarditis, studies of safety databases have reported other potential ICI-associated cardiotoxicity, including cardiomyopathy, conduction defects (heart block), atrial and ventricular arrhythmias, and pericarditis/pericardial effusions [18]. However, incidences of these other ICI-associated cardiotoxicities are rarely reported and limited to mostly case reports or case series [14]. Large population-based epidemiology studies are limited [19].Using data from a large network of health care organizations, this study aims to: (1) provide an estimate of the incidence of ICI-associated cardiotoxicity, (2) to determine patient and clinical characteristics associated with the risk of developing ICI-associated cardiotoxicity, and (3) to assess the overall survival of patients experiencing ICI-associated cardiotoxicity compared to patients who did not develop cardiotoxicity after ICI use. Besides the large sample size (5518 patients), this study used newer data that cover multiple cancer sites and all seven FDA-approved agents.Data were from health care organizations (HCOs) in the research network of TriNetX. TriNetX (www.trinetx.com; accessed on 17 February 2022) is a global federated health research network that provides real-time access to data from electronic medical records (EMRs), including demographics, diagnoses, procedures, medications, laboratory testing, vital signs, and genomic information. The network consists of academic medical centers, community hospitals, and physician practices. A PubMed search on 10 January 2022 found 119 publications using TriNetX databases. More details on how TriNetX standardizes data from contributing HCOs can be found in Harrison et al. (2020) [20]. Our institution is a contributing member of the network. We queried the database on 28 October 2019 via TriNetX’s browser-based interface of all patients with a record for ICIs. At the time of the query, there were 22 HCOs in the network with relevant data. We received a de-identified patient-level dataset for this analysis. Since the dataset is de-identified, our institution’s Internal Review Board (IRB) determined that this study is not a human-subject study. (IRB #263203).Patients who received at least one treatment with an ICI (CTLA-4 inhibitors: ipilimumab (Yervoy); PD-1 inhibitors: nivolumab (Opdivo), pembrolizumab (Keytruda), and cemiplimab (Libtayo); and PD-L1 inhibitors: avelumab (Bavencio), atezolizumab (Tecentriq), and durvalumab (Imfinzi)) by 28 October 2019 were included in the study. Users of ICIs were identified using medication and procedure files. The specific ICI agents administered during visits to HCOs were coded in the Procedure file using the Healthcare Common Procedure Coding System (HCPCS) codes. Produced by the Centers for Medicare and Medicaid Services (CMS), HCPCS is a collection of standardized codes that represent medical procedures, supplies, products, and services [21]. The medication file includes medications ordered, prescribed or administered to a patient, including mediations reported by patients in the medication list of the EMR. We searched for the specific ICI agents by name in the medication file. To ensure these medications were administered (rather than just ordered or prescribed), we required either a procedure code for specific ICI agents, or a general procedure code for chemotherapy administration in the procedure file on the same day. Therefore, we ascertained the ICI administration by (1) ICI-related HCPCS codes in the procedure file and/or (2) ICI agents identified from the medication file accompanied by a CPT/HCPCS code for chemotherapy administration or HCPCS code J9999 (antineoplastic drugs, not otherwise classified) in the procedure file on the same day. (See Supplemental Table S1 for the list of procedure codes used). This resulted in 8664 patients. The date of the first ICI administration was defined as the index date. From this cohort, we further applied the following exclusion criteria: (1) missing information on birth year or gender; (2) inconsistent information on age at death; (3) no encounters within 1 year before the index date and no encounters after the index date; (4) no diagnosis codes before or on the index date; (5) no documented non-benign cancer diagnoses before the index date to 30 days after the index date; (6) had relevant cardiac adverse event (CAE) codes prior to or on the index date (see next section on CAE for details). The final sample included 5518 patients (see Figure 1 for patient selection flowchart).ICI treatment groups were defined based on the first ICI agent(s) used by each patient. We studied these treatment groups by specific agents (nivolumab (reference group), ipilimumab, pembrolizumab, cemiplimab, avelumab, atezolizumab, durvalumab, or combination therapy), as well as by class (PD-1 inhibitors (reference group), PD-L1 inhibitors, CTLA4-inhibitors, or combination therapy). Nearly all combination therapies were for nivolumab plus ipilimumab. One patient used pembrolizumab plus ipilimumab and one used pembrolizumab plus nivolumab. These combinations were not FDA-approved therapies and may have been used in unique clinical scenarios. Therefore, we excluded these two patients from the analyses.The primary outcome was the ICI-associated cardiotoxicity defined as CAEs diagnosed within one year after ICI initiation. Patients were followed from the index date onward until 12 months, death, or the patient’s last encounter date in the database, whichever occurred earlier, to observe any CAEs. Potential CAEs were defined as new diagnoses of arrhythmia, acute myocardial infarction (AMI), myocarditis, cardiomyopathy, or pericarditis. The list of ICD-9 and 10 codes used to identify these potential CAEs were from Cathcart-Rake et al., 2020 [22] According to the authors, these conditions and corresponding codes were derived after a thorough review of irAEs noted in immunotherapy clinical trials, chemotherapy clinical trials, and case reports [22]. To reduce the risk of misclassification due to preexisting conditions, only new diagnoses of these conditions were considered and patients who had a relevant diagnosis code from the list before ICI initiation were excluded. Time to first CAE within one year after ICI initiation was calculated for patients who experienced these events. For patients who did not develop CAEs within one year after ICI initiation, their time to the first CAE was censored at 12 months, death, or their date of last contact in the database, whichever occurred first.The secondary outcome was death from any causes during the study period. To protect patients’ privacy, the de-identified data set we received included only patients’ ages at death without the exact dates of death. A patient was considered to have died if the age at death was reported. If a patient died, the last encounter date was used to impute for the date of death. Time from ICI initiation to death was analyzed. For patients who did not die during our study period, their time to death was censored at the date of last contact in the database.In addition to cardiotoxicity, ICIs may cause irAEs in other organ systems. We determined the occurrence of irAEs in other major organ systems (hematologic (anemia, thrombocytopenia, leukopenia), pneumonitis, endocrine (hypothyroidism, hyperthyroidism, hypophysitis/PGA, Hyper/hypoparathyroidism, diabetes type I, dysfunctional uterine bleeding/infertility), renal (acute kidney injury/AKI), neurological (encephalitis/myelitis/encephalomyelitis, neuritis, meningitis), hepatic (hepatitis), gastrointestinal (GI) (colitis, pancreatitis, mucositis), and skin (vitiligo)) within one year of ICI initiation. The list of ICD-9 and 10 codes used to identify these potential CAEs were from Cathcart-Rake et al., 2020, who developed these codes from a thorough review of irAEs noted in immunotherapy clinical trials, chemotherapy clinical trials, and case reports [22].Patient demographic and clinical characteristics included age on the index date, gender, ethnicity, race, comorbidities, primary cancer site, type of first ICI treatment, and any prior cancer treatment (chemotherapy, radiation). Cancer sites included all sites with an FDA-approved ICI-indication (melanoma, lung, renal cell carcinoma (RCC), urothelial (urethra/bladder/ureter), Hodgkin’s lymphoma, Merkel cell carcinoma (MCC), gastric, colon, breast, cervical, primary mediastinal (thymic) large B-cell lymphoma), and other (include all other cancer sites) based on diagnoses codes before ICI initiation to 30 days after. Comorbidity burden was measured using the Deyo–Charlson comorbidity index based on diagnoses reported prior to or on the index date. We used the enhanced ICD-9 codes and ICD-10 codes developed and validated for consistency by Quan, 2005 [23]. The enhanced ICD-9 codes were more consistent with the ICI-10 codes and performed slightly better than the original ICD-9 codes in predicting in-hospital mortality [23]. Diagnoses of neoplasms were not included in the calculation of the index since all our subjects had cancer. However, metastatic/secondary cancers were included. Hierarchy coding was applied to prevent duplicated accounting: if a person had diagnoses of both a mild and a severe form of the disease prior to ICI use (e.g., mild and moderate/severe liver disease, diabetes with and without chronic complications), the patient was only scored on the more severe disease in the CCI [24]. Chemotherapy included all systemic agents (oral chemotherapy agents were not considered) identified using HCPCS codes in the procedure file. Radiation therapy included any radiation therapy identified using procedure codes for radiation treatment delivery (CPT/HCPCS codes) in the procedure file. Stage information was only available for a small subset of patients and, therefore, was not included in the analysis. However, during the study period, ICIs were approved to treat advanced cancers and, in our study cohort, 86% had metastatic diagnoses prior to or on the index date.Patients’ demographic and clinical characteristics were summarized using frequencies and means (standard deviations). These characteristics were compared between patients who developed CAEs within one year post ICI initiation and those who did not using t-tests for continuous variables and Chi-square tests for categorical variables. These characteristics by first ICI agent(s) and by primary cancer sites were summarized using frequencies (Supplemental Tables S2 and S3). We calculated the proportion of patients who developed any CAEs as well as proportions of patients who developed each CAE during the 12 months after ICI initiation. To account for differences in follow-up periods, we also reported the rates of CAEs per person/year and their 95% confidence intervals (95% CI), calculated using the quadratic approximation to the Poisson log likelihood for the log-rate parameter [25].Time to first CAE within one year after ICI initiation was analyzed using survival analysis. Although Kaplan–Meier (KM) method and Cox proportional hazard (PH) models are often used to estimate survival and associated hazard ratios (HRs), these methods treat death as an uninformative censoring event [26]. In our study population of patients with advanced stage cancers, the mortality rate was high (29.1% died within one year of ICI initiation). This creates competing risk for CAEs (i.e., a patient died before experiencing CAEs). In the presence of competing risk, the KM method overestimates the cumulative incidence rate [26]. Therefore, we estimated cumulative incident rates and adjusted hazard ratios (aHRs) of CAEs using the competing-risks regression model according to the method of Fine and Gray (1999) [27]. Use of the Fine–Gray sub-distribution hazard model is recommended when the focus is on estimating incidence or predicting prognosis in the presence of competing risks [26]. “Failure to account correctly for competing events can result in adverse consequences, including overestimation of the probability of the occurrence of the event and misestimation of the magnitude of relative effects of covariates on the incidence of the outcome” [26]. Two regression analyses were conducted to assess the association with: (1) different ICI agent(s); and (2) different ICI classes (PD-1 inhibitors, PD-L1 inhibitors, CTLA4-inhibitors, or their combination (ipilimumab + nivolumab)). Both models adjusted for differences in patient demographics (age, gender, race, ethnicity), comorbidity index, pre-existing cardiovascular conditions (hypertension, cerebrovascular disease (CED), congestive heart failure (CHF), myocardial infarction (MI), peripheral vascular disease (PVD)), renal disease, moderate/severe liver disease, and major (≥30 cases in CAE and non-CAE groups) cancer sites with FDA indication for ICI use by 2018 (lung, melanoma, RCC, urothelial, head and neck, MCC, and liver). Although MCC is a rare cancer, its mortality is high [28]. We included MCC as a covariate so that it would not bias the estimates of the major cancer sites when compared to other cancers. Liver cancer was also included because the proportion of patients with liver cancer was statistically significant between the CAE and non-CAE groups in the bivariate analysis. After each regression, we presented the adjusted cumulative incident curves by first ICI agent(s) and ICI class, respectively.For overall survival, we used KM survival curves to summarize time to death due to all causes after ICI initiation. The aHRs from a multivariate Cox’s PH regression model were used to compare overall survivals between patients who developed CAEs within one year against those who did not. For this analysis, a binary variable indicating whether a patient developed CAEs within one year after ICI initiation (Y/N) was included as a covariate in addition to the aforementioned covariates. Considering the Schoenfeld residuals-based test [29], which showed a significant violation of the PH assumption (p = 0.0033), we decided to use a time-varying Cox regression model by adding an interaction term of time with the indicator variable, the coefficient of which revealed how the HR changed over time between patients who developed CAEs within one year and those who did not. As previously described, myocarditis is a rare but highly fatal CAE. For patients who developed myocarditis within one year and died afterwards (n = 55), we also graphically presented the time to first myocarditis diagnosis and time to death after myocarditis diagnosis for each patient to graphically show the distributions of each variable and examine if they were correlated (i.e., whether time to death after myocarditis depended on how soon a patient developed myocarditis after ICI initiation). We used SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and Stata version 17 (Stata Corp, College Station, TX, USA) for all analyses.After applying the inclusion and exclusion criteria, the final data set included 5518 patients who received at least one cycle of ICI treatment and had no prior documented cardiotoxicity codes. Among them, 691 developed CAEs within one year of ICI initiation (12.5%) (Figure 1).Table 1 reports patients’ demographic and clinical characteristics. Compared to patients who did not develop CAEs within one year of ICI initiation, patients who developed CAEs were older (Age ≥ 65 years 54.6% vs. 44%, p < 0.0001), more likely to be males (63.2% vs. 56.7%, p = 0.0012), less likely to be Hispanic (1.4% vs. 3.4%, p = 0.0144), had higher comorbidity burdens (p = 0.0042), and were more likely to have pre-existing cardiovascular diseases (CED: 12.9 vs. 10.3%, p = 0.0375; CHF: 5.9% vs. 2.6%, p < 0.001; MI: 4.8% vs. 2.3%, p = 0.0002; PVD, 15.6% vs. 12.3%, p = 0.0129; and hypertension, 50.8% vs. 45.1%, p = 0.0046) or renal disease (RD: 13.7% vs. 10.8%, p = 0.0221)). Lung (35.7%) and melanoma (33.7%) were the most common cancer sites. Patients who developed CAEs within one year of ICI initiation were more likely to have lung cancer (42.3% vs. 34.7%, p < 0.0001) and less likely to have liver cancer (2.8% vs. 6.6%, p < 0.0001) compared to those who did not. About one third of patients received some radiation treatment and 22% had received chemotherapy prior to ICI initiation. There were no differences in these prior treatments between patients who developed CAEs within one year of ICI initiation and those who did not (p > 0.05). The most common chemotherapy agents used were alkylating agents (84% of patients who received any chemotherapy prior to ICI initiation). Presence of irAEs in other organ systems during the 12 months after ICI initiation was high, ranging from 7.6% for skin to 44.9% for endocrine. Patients who developed CAEs within one year were also more likely to experience those other irAEs compared to patients who did not (all p < 0.05), except for skin irAEs (p = 0.3160).Table 2 reports the ICI treatment received by patients. Patients who developed CAEs within one year of ICI initiation were more likely to have received ipilimumab monotherapy (12.3% vs. 8.0%) as their first ICI treatment whereas those who did not develop CAEs were more likely to have initiated on nivolumab monotherapy (28.4% vs. 32.5%) or its combination with ipilimumab (6.9% vs. 8.7%). The overall difference in the first ICI treatments between the two groups was statistically significant (p = 0.0038). During the study period, the majority (84%) received treatment with only one ICI agent but 16% received two or three different ICI agents (switched or used in combination therapies) during the study period. There was no difference in the number of ICI agents used between patients who developed CAEs within one year after ICI initiation and those who did not (p = 0.6878).During the 12 months after ICI initiation, 691 (12.5%) patients developed CAEs. After adjusting for differences in follow-up, the rate was estimated to be 0.20 (95% CI: 0.19–0.22) per person/year. The most common CAEs were arrhythmia (9.3%) followed by myocarditis (2.1%), acute MI (1.7%), pericarditis (1.2%), and cardiomyopathy (0.9%). Estimated rates per person/year for each type of CAE are reported in Supplemental Table S4. Table 3 presents the aHRs estimated from competing risk regressions by first ICI agent(s) and by ICI class. The aHRs for covariates were very similar in both models. Here, we only report the results from the model by first ICI agent(s). Risk of CAEs increased with age. Patients aged 65–74 years (aHR: 1.45; 95% CI: 1.03–2.05; p = 0.035) and 75 years or older (aHR: 2.11; 95% CI: 1.47–3.03; p < 0.001) at the time of ICI initiation were statistically significantly more likely to develop CAEs within one year compared to patients aged 18–44 years. Males (aHR: 1.31; 95% CI: 1.12–1.54; p = 0.001) and blacks (aHR: 1.34; 95% CI: 1.02–1.77; p = 0.038) had higher risk of developing CAEs compared to females and whites, respectively. Patients with prior diagnosis of CHF (aHR: 2.01; 95% CI: 1.43–2.83; p < 0.001) and MI (aHR: 1.50; 95% CI: 1.04–2.17; p = 0.029) had a statistically higher risk of CAEs compared to those without those prior diagnoses. Having lung cancer statistically significantly increased (aHR: 1.24; 95% CI: 1.02–1.50; p = 0.032) and liver cancer significantly decreased (aHR: 0.45; 95% CI: 0.27–0.74; p = 0.002) the risk of CAEs compared to those with other cancers.After adjusting for these differences, patients who initiated on monotherapy with ipilimumab (aHR: 2.00; 95% CI: 1.49–2.70; p < 0.001) or pembrolizumab (aHR: 1.21; 95% CI: 1.01–1.46; p = 0.040) had a higher risk of developing CAEs within one year compared to nivolumab monotherapy. Risk of cardiotoxicity was also elevated among patients who used avelumab as their first ICI treatment (aHR: 1.92; 95% CI: 0.85–4.34; p = 0.117), although the risk was not statistically significant likely due to the small sample size (n = 29). Similarly, when compared across ICI classes, CTLA4 inhibitors (ipilimumab) had the highest risk of CAEs compared to other classes (aHR: 1.77; 95% CI: 1.35–2.34; p < 0.001). The adjusted cumulative incidence rates by the first ICI agent(s) and by ICI class are presented in Figure 2a,b.Risk of death was higher among patients who developed CAEs within one year of ICI initiation; 55% died during our study period compared to 35% in patients who did not. Figure 3 presents the survival curves accounting for censoring due to different follow-up periods. The median survival time was a little over a year after ICI initiation in patients who developed CAEs within one year of ICI initiation compared to about 2.5 years for those who did not develop CAEs (p < 0.0001) (Figure 3). Table 4 reports the aHRs from the time-varying Cox regression model. While the baseline HR was 1.46 (95% CI: 1.26–1.70; p < 0.001) between patients who developed CAEs within one year and those who did not, the HR widened by approximately 16% each year. (Table 4). Among the 116 patients who developed myocarditis within the first year (2.1%), the median time to myocarditis diagnosis was 115 days, with 25% developing it within 10 days and 75% developing it within 6 months of ICI initiation. Among them, 55 (47%) patients died afterward. While most of the deceased patients (49/55, 89.1%) died within one year after developing myocarditis, four (7%) died during the second year and only two patients lived beyond four years (3.6%) (Figure 4).In this large EMR-based database from multiple health care organizations, we determined the risk of cardiotoxicity within one year of ICI initiation among patients who had received at least one cycle of ICI for multiple cancers. Similar to other large population-based studies using real-world data, we found a higher incidence of cardiotoxicity than previously reported in clinical trials. By month 12, 12.5% developed cardiotoxicity. The most common cardiotoxicity was arrhythmia (9.3%) and 2.1% developed myocarditis by month 12. Cardiotoxicity is often underreported in clinical trials. Hu et al. (2017) reviewed 22 clinical trials involving single-agent PD1 and PD-L1 inhibitors in non-small-cell lung cancer (NCSLC). Overall, cardiotoxicity was reported in 12 patients (one myocarditis, one pericardial effusion, one cardiac tamponade, one pulmonary embolism, one constrictive pericarditis, two MI, two cardiorespiratory arrest, and three cardiac (heart) failure) out of a combined 1784 patients across 10 trials that reported any cardiotoxicity (0.7%) [30]. Our estimate is higher than that reported in clinical trials but similar to other studies using real world databases. Using a US commercial insurance database (OptumLabs Data Warehouse, https://www.optum.com/about-us.html, accessed on 17 February 2022), Cathcart-Rake et al. (2020) evaluated irAEs incidence in 3164 patients (https://www.optum.com/about-us.html, accessed on 17 February 2022) with NCSLC who received PD-1 or PD-L1 inhibitors between 2015 and 2017 [22]. By month 9, 9.07% experienced an arrhythmia, 2.85% had acute MI, 0.89% had myocarditis, 1.65% had pericarditis, and 1.02% had cardiomyopathy [22]. A nationwide Danish study using data from 2011 to 2017 examined a composite outcome of cardiac events (arrhythmia, pericarditis, myocarditis, and heart failure) or cardiovascular death [31]. The one-year absolute risk of cardiac events after ICI initiation was 6.6% (95% CI: 1.8–11.3) in melanoma patients treated with PD-1 inhibitors, 7.5% (3.7–11.3) in melanoma patients treated with CTLA-4 inhibitors, and 9.7% in lung (95% CI: 6.8–12.5) cancer patients treated with PD-1 inhibitors [31]. Compared to patients without ICI treatment, the risk of cardiac events was higher in patients treated with ICIs but decreased after 6 months [31]. Chitturi et al. (2019) studied 252 patients with pathologically confirmed lung cancer who received ICIs between August 2015 and August 2018 from a single institution [32]. In this study, major adverse cardiac events, defined as a composite of cardiovascular death, nonfatal infarction, nonfatal stroke, and hospitalization for heart failure, occurred in 13.3% of patients during a median follow-up of 6 months with a median time to event of 51 days [32]. In another one-institution study of 424 cancer patients who received any ICI treatment from 2011 to 2017, 14.6% developed cardiovascular diseases after initiation of ICI treatment, defined as a new ICD diagnosis code for cardiomyopathy, heart failure, arrhythmia, heart block, pericardial disease, or myocarditis [33]. Similar to our study, the most frequently diagnosed cardiac condition was arrhythmia (6.1%) and 5.4% of patients had newly diagnosed heart failure [33]. However, the aforementioned studies used data of earlier years (mostly from 2017, [22,31,33]), from a single institution [32,33], or focusing on one or two cancer sites [22,31,32]. We provided updated information with newer data and a much larger cohort across all cancer sites and all seven FDA-approved ICIs and combination therapies. Mortality in ICI treated patients who developed cardiotoxicity was higher compared to those who did not. In our study, we found that 55% of patients who developed CAEs within one year of ICI initiation died compared to 35% in those who did not. This finding is consistent with previous studies. In a single-institution study of 424 cancer patients who received any ICI treatment from 2011 to 2017, 66.1% of patients with a concomitant diagnosis of incident cardiovascular disease died compared to 41.4% among those who did not (odds ratio (OR): 2.77; 95% CI: 1.55–4.95; p  =  0.0006) [33]. Escudier et al. reported a fatality rate of 27% among 30 patients with ICI-associated cardiotoxicity including left-ventricular systolic dysfunction, Takotsubo syndrome-like appearance, atrial fibrillation, ventricular arrhythmia, conduction abnormalities, and pericardial effusions [17]. Of the 122 ICI-associated myocarditis cases, 20 pericardial disease cases, and 82 vasculitis cases identified in the WHO’s global database of individual case safety reports, 50% of myocarditis cases, 21% of pericardial disease cases, and 6% of vasculitis cases resulted in death, respectively [13]. Previous studies of clinical trial data have found elevated risk for severe cardiotoxicity when an ICI was used in combination (nivolumab plus ipilimumab) or with chemotherapy. For instance, Hu et al. (2021) conducted a meta-analysis of cardiac adverse events in 20,244 patients from 25 clinical trials involving monotherapy or combination therapy of ICIs plus chemotherapy published up to October 2020 [34]. Cardiac adverse events were classified into six major categories: arrhythmias, cardiac failure, coronary artery disease, pericardial disease, cardiac arrest, and myocardial disease [34]. Compared with nivolumab or ipilimumab monotherapy, combined nivolumab and ipilimumab therapy showed significant increases in grade 5 arrhythmias (OR: 3.90; 95% CI: 1.08–14.06) where arrhythmias was defined broadly to include atrial fibrillation, atrial flutter, atrial tachycardia, atrioventricular block, arrhythmia supraventricular, complete atrioventricular block, bradycardia, bifascicular block, sinus bradycardia, sinus tachycardia, supraventricular tachycardia, tachycardia, ventricular arrhythmia, ventricular tachycardia, and ventricular fibrillation [34]. In our study, use of the nivolumab plus ipilimumab combination as the first ICI treatment trended towards a higher risk of CAEs compared to nivolumab monotherapy but did not reach statistical significance after adjusting for differences in patient characteristics. On the other hand, patients initiated on ipilimumab, pembrolizumab, and avelumab monotherapies were found to have an elevated risk compared to nivolumab monotherapy, although avelumab did not achieve statistical significance due to small sample size. These results are not directly comparable to Hu et al. (2021) because the severity of cardiotoxicity could not be ascertained in our study. In our study, there were no differences in prior radiation and chemotherapy use between the CAEs group and no CAES group. Adjusting for prior use of these treatments or the chemotherapy agents did not estimate a statistically significant association with CAEs either. We therefore did not include them in the reported analysis. In the above meta-analysis by Hu et al. (2021) using clinical trial data, PD-1 inhibitor plus chemotherapy showed a significant increase in grade 1–5 myocardial disease (OR: 5.09; 95% CI: 1.11–23.32) compared with chemotherapy alone [34]. Compared with combined chemotherapy and nivolumab/ipilimumab, combined nivolumab and ipilimumab therapy showed a significant increase in grade 1–5 arrhythmias (OR: 2.49; 95% CI: 1.30–4.78) [34]. However, a study of ICI use in elderly patients with lung cancer using the SEER-Medicare database found that the ICI-plus-chemotherapy (used either concurrently or sequentially) group had equal or lower risk of cardiotoxicity, including acute coronary syndrome (HR: 0.82; 95% CI: 0.64–1.05; p = 0.10), heart failure (HR: 0.74; 95% CI: 0.62–0.88; p = 0.0007), cardiac arrhythmia (HR: 0.72; 95% CI: 0.63–0.82; p < 0.0001), and heart blocks (HR: 0.48; 95% CI: 0.30–0.76; p < 0.0001), compared to the traditional chemotherapy treatment only group [35]. Given these contracting results, future studies are warranted to further investigate this in large population-based real world databases. We also found older age, male gender, black race, and history of CHF and MI were associated with increased risk of developing new CAEs within one year of ICI initiation. In the general population, risk of cardiovascular diseases (CVD) increases with age and is higher among blacks compared to other races [36]. Incidence of CVD is also lower in women than in men, although women have a higher mortality and worse prognosis after acute cardiovascular events [37]. Previous literature found increased risk of ICI-induced cardiotoxicity in males [15]. However, the findings are inconclusive regarding whether patients with pre-existing cardiovascular diseases are at increased risk of ICI-induced cardiotoxicity [1]. On the other hand, the occurrence of irAEs is often an indicator of ICI activity [38]. In a comprehensive review of existing evidence on the involvement of sociological factors, lifestyles, and metabolic disorders in modulating the ICI response in cancer patients, Deshpande et al. (2020) reported evidence on direct or indirect links of age, sex, race, lifestyle factors (diet, exercise, alcohol, and smoking), obesity, and psycho-emotional stress with ICI response; however, the findings on the selective benefits of ICI by patient’s sex or race are conflicting [39]. These findings and ours underscore the need for consideration of these factors when prescribing ICIs. The data on ICI-associated cardiotoxicities focus mostly on the development of myocarditis [40]. While clinical trials have reported a low rate of myocarditis (0.09%), real-world data have reported a higher rate. In a one-institution study of 964 patients treated with ICI from 2013 to 2017, 11 (1.14%) patients developed ICI-associated myocarditis [12]. Cathcart-Rake (2020) evaluated irAEs incidence in patients with NSCLC who received PD-1 or PD-L1 inhibitors using a US commercial insurance database and found the rate of myocarditis to be 0.89% by month 9 after ICI initiation [22]. Using the same set of diagnosis codes for cardiac irAEs, we found that 2.1% developed myocarditis within one year after ICI initiation across multiple cancers and all seven FDA-approved ICIs and ICI combination therapies. Although rare, myocarditis is often fatal. In the 18 patients who developed severe myocarditis from the Bristol Myers Squibb corporate safety database, six (33%) died [11]. Patients treated with a combination of nivolumab and ipilimumab experienced a higher incidence of myocarditis (0.27% vs. 0.06%) and fatality rate compared to nivolumab monotherapy (5/8, 63% vs. 1/10, 10%) [11]. In our study, among the 116 patients who developed myocarditis within one year, 47% died afterwards with nearly 89% of them having died within one year of myocarditis diagnosis. In 122 cases of ICI-associated myocarditis identified from the WHO’s global adverse report database of individual case safety reports, death occurred in 50% of the cases [13]. Despite the known risk, ICI-induced myocarditis is still poorly understood. Pathological studies have demonstrated heart injury from T-cell infiltration within myocardium with or without myocyte degeneration and necrosis of non-ischemia origin [1,41]. A recent study by Power et al. (2021) reported electrocardiographic and arrhythmogenic features of ICI-myocarditis among 125 patients identified from an online registry of 49 institutions and 11 countries [42]. The results from this study establish ICI-myocarditis to be highly arrhythmogenic and define specific electrocardiographic features that will help clinicians diagnose and prognosticate the syndrome [42]. A wide range of ECG abnormalities have been presented, including conduction blocks, decreased voltage, and repolarization abnormalities that frequently degenerate to malignant arrhythmias [42]. Currently, endomyocardial biopsy remains the gold standard for confirming myocarditis, but it is rarely performed in clinical practice due to its invasive nature. In our study and others [33], arrhythmias were found to be the most commonly diagnosed CAEs, some of which could be due to undiagnosed underlying myocarditis. Although ICI-associated myocarditis typically occurs after 2–3 ICI cycles, a wide range of onset times have been reported, from 2 to 454 days after starting ICI treatment [1,11,12,17,43]. Our study findings are consistent with these reports. Among the 116 patients who developed myocarditis within one year, the median time to onset was 115 days with 25% developing within 10 days and 75% within 6 months. Some late-onset of ICI-associated myocarditis occurred more than a year after starting ICI therapy [44], although it is unclear whether it was due to delayed development of myocarditis or resulted from myocarditis that began much earlier, or was caused by cumulative injury to the heart due to persistent systemic immune activation and inflammation [1]. Occupation of PD-1 and PD-L1 receptors may remain long after the infusions of ICI have stopped, which may partially explain the wide range in the median time to onset of myocarditis and requires clinicians to remain vigilant when patients present with myocarditis-like symptoms late after starting ICI or who are no longer being actively treated with an ICI [1]. Most studies of cardiotoxicity focused on earlier ICIs such as ipilimumab, nivolumab, pembrolizumab, or their combinations. Use of ICI combination therapy increases the risk of myocarditis compared to monotherapy [11]. Few studies have reported on cardiotoxicity associated with the newer ICIs, most of which are PD-L1 inhibitors. It was speculated that PD-L1 inhibitors may be associated with lower adverse events because they still allow for the interaction of PD-1 with its other ligand PD-L2 [45]. However, a systematic review of published data of trials utilizing PD-1 (nivolumab and pembrolizumab) and PD-L1 inhibitors (atezolizumab, durvalumab, and avelumab) in NSCLC patients found similar toxicity between PD-1 and PD-L1 inhibitors [45]. A significantly higher (but not reaching statistical significance) rate of toxicity was observed in durvalumab compared to other ICIs (75% vs. 62–67%), which warrants future studies [45]. Cardiotoxicity was not separately studied in this study [45]. In our study, we found that after adjusting for patients’ demographic and clinical characteristics, patients initiated on ipilimumab and pembrolizumab monotherapies had a statistically significant higher risk of developing CAEs within one year after ICI initiation compared to patients initiated on nivolumab monotherapy, with the ipilimumab group having nearly double the hazard of CAEs compared to the nivolumab group. Patients initiated on avelumab monotherapy also were estimated as having a highly elevated risk but the effect was not statistically significant due to the small sample size. Larger studies are needed to confirm this finding. We did not find statistically significant differences in CAE risk between other PD-L1 agents (atezolizumab and durvalumab) and nivolumab monotherapy, nor did we find statistically significant difference by class between PD-L1 and PD-1 inhibitors when comparing patients’ initial ICI treatment. Few studies have compared ICI-associated cardiotoxicity across cancer sites. In a recently published nationwide Danish study using data from 2011 to 2017, patients with incident lung cancer or melanoma were studied [31]. Cardiotoxicity was defined as a composite outcome of cardiac events (arrhythmia, pericarditis, myocarditis, or heart failure) or cardiovascular death. The one-year absolute risk of cardiac events was 6.6% (95% CI: 1.8–11.3) in melanoma patients treated with PD-1 inhibitors, 7.5% (3.7–11.3) in melanoma patients treated with CTLA-4 inhibitors, and 9.7% in lung (95% CI: 6.8–12.5) cancer patients treated with PD-1 inhibitors [31]. In addition, lung cancer and melanoma patients were studied separately; thus, it is unknown whether these differences were statistically significant [31]. Moreover, higher proportions of lung cancer patients have pre-existing cardiovascular conditions prior to ICI initiation compared to melanoma patients [31]. In the adjusted analysis after adjusting for differences in patient characteristics, melanoma patients with ICIs (vs. melanoma patients without ICIs) were estimated to have a much higher HR of cardiac events compared to lung cancer patients with ICIs (vs. lung cancer patients without ICIs) (e.g., for risk of cardiac events occurring <6 months after ICI initiation: melanoma with PD-1 inhibitor (HR: 4.30; 95% CI: 1.38–13.42), lung cancer with PD-1 inhibitor (HR: 2.14; 95% CI: 1.50–3.05)) [31]. However, these HRs could not be directly compared between melanoma and lung cancer patients because the reference groups were different (melanoma patients without ICIs and lung cancer patients without ICIs) [31]. Waheed et al. compared the primary cancer diagnosis between patients with newly diagnosed cardiovascular disease and those without and find no statistically significant difference; no further adjusted analyses by cancer sites were conducted [33]. Two studied only lung cancer patients [22,32]. These studies used different designs and definitions of cardiotoxicity and could not be compared directly [22,31,32,33]. In our study, we included ICI users with malignant cancers in multiple sites and found that having lung cancer independently increased the risk of ICI-associated cardiotoxicity compared to those without lung cancer. Prior thoracic radiation treatment for lung cancer can cause injury to the heart. Animal studies have characterized radiation-induced heart disease with fibrosis and acute production of inflammatory cytokines, which can compound ICI-induced cardiac dysfunction and cause cumulative cardiotoxicity [46]. Interestingly, in a recent review study of 134 published cardiotoxicity cases, lung cancer appeared to have a longer time of onset of cardiotoxicity comparing to other cancer sites [18]. The reason for this difference is unknown and should be further investigated in future larger studies. We also found liver cancer to be associated with statistically significantly lower risk of ICI-associated cardiotoxicity compared to other cancer sites. To the best of our knowledge, no studies have compared ICI-associated cardiotoxicity between liver cancer and other cancers. The reason for this reduced risk is unknown and warrants future studies. In a large study of health insurance claims database, Wang et al. found that although ICIs were associated with increased risk of developing irAEs in patients with all seven cancer types under study compared to chemotherapy, the risk varied across the cancer types; however, cardiac irAEs were not included in the analysis [47]. Although ICIs were initially approved as salvage treatments when patients failed other treatments, they are increasingly being approved for earlier stage cancers and in adjuvant settings. These patients may survive longer and are at more risk of late-onset ICI-associated cardiotoxicity; thus, continued vigilance is needed. The most common irAEs are endocrine AEs, with 44.9% patients experiencing those AEs in the first year after ICI initiation. This estimate is consistent with previous studies. A review of phase-III studies of ICIs found varying estimates ranging from 3.8% for nivolumab in one study to up to 30% in nivolumab and ipilimumab combinations for endocrine AEs of any grade; however, high-grade (grade 3–4) endocrine AEs were rare (0–5.5% across studies) [48]. AKI was also high in our study with 22.1% of patients having developed AKI within one year of ICI initiation. AKI occurred in 35.3% of patients who developed CAEs within one year of ICI initiation and 20.2% in those who did not. The difference was statistically significant (p < 0.0001). Because the focus of this study was on cardiac AEs, we did not exclude patients with pre-existing AKIs or renal disease. In this study, 11.2% had pre-existing renal disease. The proportion of patients with pre-existing renal disease was higher in patients who developed CAEs within one year of ICI initiation than those who did not (13.7% vs. 10.8%, p = 0.0221). We adjusted for this difference in the regression analysis. In a study of lung cancer patients, Cathcart-Rake et al. (2020) excluded patients with pre-existing codes for AKI and other irAEs, and the estimated incidence rate of AKI by month 9 was 7.33% (95% CI: 6.18, 8,69) [22]. A review of published phase-2 and -3 clinical trials found the overall incidence of AKI to be 2–5%, with high-grade AKI (grade 2 or 3) needing dialysis to be 0.6% [49]. However, other studies using routine practice data have reported higher incidences of 13.9% [50] to 29% [51], varying by ICI agent and higher in the combination therapy [52].This is a retrospective study using real world data from a network of multiple health care systems. The major advantages are its large sample size (5518 patients), newer data, and its covering of multiple cancer sites and all seven FDA-approved agents, although the number of records for newer agents (atezolizumab, avelumab, cemiplimab, and durvalumab) are relatively small (21 to 201 patients depending on the agent). A major limitation is that tumor-specific characteristics such as stage and histology were not included due to low reporting to the network. Moreover, severity of cardiotoxicity could not be determined. ICI-related cardiotoxicity was defined based on diagnosis codes, which may be subject to inaccurate reporting. To reduce the risk of misclassification due to pre-existing conditions, only new diagnoses were considered and patients who had a relevant diagnose code from the list of CAEs before ICI initiation were excluded. We relied on this temporal relationship (not present before ICI initiation but present within one year after ICI initiation) to establish incident cases, which is a common epidemiology study design. We further restricted the study to only cases discovered within one year of ICI initiation to reduce the chances of these events occurring later due to the natural aging process, other drugs, or newly developed conditions unrelated to ICI use. Nonetheless, it is possible that some cases may have been misclassified and late-onset cases may have been missed due to this cutoff of 12 months. Moreover, cardiotoxicity was limited to the list of CAEs examined. Recent studies have found associations between ICIs and increased thromboembolic events [53] and atherosclerotic plaque [40]. Therefore, we may have underestimated the incidence of ICI-associated cardiotoxicity.Thoracic radiation therapy has been shown to increase the risk of cardiotoxicity due to its proximity to the heart. However, radiation fields could not be determined from the procedure codes. Nonetheless, the analysis adjusted for lung cancer diagnosis, which is the major site for ICI indication and thoracic radiation. Other cancer sites such as breast and esophagus may also receive thoracic radiation. However, by October 2019, only atezolizumab had been approved for triple-negative breast cancer (March 2019) and pembrolizumab in combination with chemotherapy was approved for esophageal cancer as a third or subsequent line of treatment (September 2017) [2]. The use of ICIs for patients with these types of cancer was very low during our study period and, therefore, was not separately analyzed (Table 1). As a retrospective study, it is possible that unmeasured residual confounders such as diet, physical activity, and family history may influence the association between ICI use and CAEs [39]. We also did not include smoking status and BMIs because not all HCOs reported this information to TriNetX. However, previous studies that included smoking status and/or BMI did not find a significant effect of either variable on ICI-associated cardiotoxicity [40,48]. Information on tumor PD-L1 proportion and mutation burden are very incomplete and, therefore, were not used in the analysis. Although both affect responses to ICIs, it is unknown whether they affect the risk of ICI-associated cardiotoxicity. We also did not assess the association of biomarkers with ICI-associated cardiotoxicity due to incomplete information. Biomarkers such as troponin and B-type natriuretic peptide are often elevated in patients with ICI-associated myocarditis [12] and can aid in diagnosis and assessing prognosis [54]. However, it is still unclear whether these biomarkers could be used to identify high-risk patients who will develop myocarditis [55].In this large EMR-based database from multiple health systems, we estimated the incidence rates of CAEs within one year after ICI initiation. While there were no differences in risk of cardiotoxicity between PD-1 and PD-L1 inhibitors overall, ipilimumab and pembrolizumab use may increase the risk of cardiotoxicity compared to other agents and should be closely monitored in the future, especially with the rapidly expanded use of pembrolizumab. Avelumab was also estimated as having a highly elevated risk compared to nivolumab and other PD-L1 agents, although the estimate did not reach statistical significance. Given that little is known on the cardiotoxicity among PD-L1 agents or avelumab, future larger studies are urgently needed to confirm this finding.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051145/s1, Table S1. Procedure codes for ICIs; Table S2. Patient Characteristics by First ICI treatment; Table S3 Patient Characteristics by Primary Cancer Site; Table S4. Estimated Rate Per Person-Year.Conceptualization, C.L.; methodology, C.L., S.A.B. and J.Y.; software, C.L.; validation, C.L., S.A.B. and J.Y.; formal analysis, C.L.; investigation, C.L., S.A.B. and J.Y.; resources, C.L.; data curation, C.L..; writing—original draft preparation, C.L.; writing—review and editing, C.L., S.A.B. and J.Y. All authors have read and agreed to the published version of the manuscript.C.L. was supported by the National Institutes of Health (NIH) and the National Center for Advancing Translational Sciences (NCATS) (UL1TR003107). The views expressed here are those of the authors and do not necessarily represent any official position of the National Institutes of Health.Not applicable. This study used de-identified data and was determined not to be human-subject research by the University of Arkansas Medical Sciences Internal Review Board (IRB #263203).Not applicable.Data are proprietary and cannot be shared.We thank Laura F. Hutchins for reviewing the study’s findings and providing feedback.All authors declare no conflicts of interest. C.L. received research funding for an unrelated project sponsored by the University of Utah/AstraZeneca.Patient selection flowchart.Cumulative incidence of cardiotoxicity after competing risk regressions, adjusted for covariates. (a) By first ICI agent(s), aHR (95% CI). Nivolumab: 1.00 (reference); atezolizumab: 1.11 (0.71–1.72), p = 0.65; avelumab: 1.92 (0.85–4.34), p = 0.12; cemiplimab: 0.64 (0.08–4.75), p = 0.66; durvalumab: 1.01 (0.47–2.16), p = 0.97; ipilimumab: 2.00 (1.49–2.70), p < 0.01; pembrolizumab: 1.21 (1.01–1.46), p = 0.04; combination (niv + ipi): 1.18 (0.85–1.64), p = 0.32. (b) By ICI Class, aHR (95% CI). PD-1: 1.00 (reference); CTLA4: 1.77 (1.34–2.34), p < 0.01; PD-L1: 1.04 (0.74–1.46), p = 0.82; combination (niv + ipi): 1.06 (0.78–1.45), p = 0.71. ICI: immune checkpoint inhibitor; CTLA-4: Cytotoxic T-lymphocyte associated-antigen-4; PD-1: programmed death receptor-1; PD-L1: programmed death-ligand 1; Combo (ipi+niv): combination of ipilimumab and nivolumab; aHR: adjusted hazard ratio; 95% CI: 95% confidence interval.Kaplan–Meier estimate of overall survival. CAE: cardiac adverse event. ICI: immune checkpoint inhibitor.Time to myocarditis diagnosis and time to death from myocarditis diagnosis in patients who developed myocarditis and died during the study period (n = 55).Patient Characteristics.ICI: immune checkpoint inhibitor; irAE: immune-related adverse event. 1 Comorbidities and Charlson Comorbidity Index (CCI) values were calculated based on diagnoses before or on the day of ICI initiation. ICD 9 and 10 codes used to calculate the CCI were from Quan et al. (2005). Primary cancer diagnoses were not included in the calculation of the index but metastatic/secondary cancers were included. Hierarchy coding was applied to prevent duplicated accounting. For instance, if a person had diagnoses of both a mild and a severe form of the disease (e.g., mild and moderate/severe liver disease, or diabetes with and without chronic complications), the patient was only scored on the more severe disease in the CCI. 2 Any diabetes included patients who had any of the following diagnoses, diabetes without chronic complication, diabetes with chronic complication, or diabetes secondary to drug use, prior to or on the day of ICI initiation. 3 Primary cancer sites are based on diagnoses of primary cancers (excluding secondary diagnoses) before ICI initiation or 30 days after. Site variables are not mutually exclusive and patients may have multiple cancer sites. 4 Other irAEs included hematologic (anemia, thrombocytopenia, leukopenia), pneumonitis, endocrine (hypothyroidism, hyperthyroidism, hypophysitis/PGA, hyper/hypoparathyroidism, diabetes type I, dysfunctional uterine bleeding/infertility), renal (acute kidney injury/AKI), neurological (encephalitis/myelitis/encephalomyelitis, neuritis, meningitis), hepatic (hepatitis), gastrointestinal (GI) (colitis, pancreatitis, mucositis), and skin (vitiligo)).First immune checkpoint inhibitor treatments.ICI: immune checkpoint inhibitor. CTLA-4: Cytotoxic T-lymphocyte associated-antigen-4. PD-1: programmed death receptor-1. PD-L1: programmed death-ligand 1. Combo: combination. Hazard ratios for risk of cardiotoxicity estimated from competing risk regressions.ICI: immune checkpoint inhibitor; aHR: adjusted hazard ratio; 95% CI: 95% confidence interval; CTLA-4: Cytotoxic T-lymphocyte associated-antigen-4; PD-1: programmed death receptor-1; PD-L1: programmed death-ligand 1; Combo (ipi+niv): combination of ipilimumab and nivolumab. 1 Charlson Comorbidity Index (CCI) values were calculated based on diagnoses before or on the day of ICI initiation. The ICD 9 and 10 codes used to calculate the CCI were from Quan et al. (2005). 2 Primary cancer sites are based on diagnoses of primary cancers (excluding secondary diagnoses) before ICI initiation or 30 days after. Site variables are not mutually exclusive and patients may have multiple cancer sites.Hazard ratios for overall survival.ICI: immune checkpoint inhibitor; aHR: adjusted hazard ratio; 95% CI: 95% confidence interval. 1 Charlson Comorbidity Index (CCI) values were calculated based on diagnoses before or on the day of ICI initiation. The ICD 9 and 10 codes used to calculate the CCI were from Quan et al. (2005). 2 Primary cancer sites are based on diagnoses of primary cancers (excluding secondary diagnoses) before ICI initiation or 30 days after. Site variables are not mutually exclusive and patients may have multiple cancer sites.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01146.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Skin toxicity is one of paclitaxel’s adverse effects. However, its real impact on the skin could be underestimated as these alterations can also appear asymptomatic. We have observed that paclitaxel modifies gene and protein expression of skin markers in a 3D epidermis model, and impairs physical, physiological, and biomechanical properties of the skin in gynecologic cancer patients. These subclinical alterations might be avoided by using prophylactic measures during treatment to prevent possible future adverse reactions.Background: Paclitaxel is a microtubule-stabilizing chemotherapeutic agent. Despite its widespread use, it damages healthy tissues such as skin. The goal of this study was to prove that the real impact of paclitaxel-induced skin toxicity could be underestimated because the adverse events might appear asymptomatic. Methods: Gynecological cancer patients were recruited. Skin parameters measurements were taken after three and six paclitaxel cycles. Measurements were conducted using specific probes which measure hydration, transepidermal water loss (TEWL), sebum, elasticity and firmness, erythema, roughness, smoothness, skin thickness, and desquamation levels. Further, a 3D epidermis model was incubated with paclitaxel to analyze gene and protein expression of aquaporin 3, collagen type 1, elastin, and fibronectin. Results: Paclitaxel induced alterations in the skin parameters with no visible clinical manifestations. Gynecological cancer patients under paclitaxel treatment had a decrease in hydration, TEWL, sebum, elasticity, and thickness of the skin, while erythema, roughness, and desquamation were increased. The molecular markers, related to hydration and the support of the skin layers, and analyzed in the 3D epidermis model, were decreased. Conclusions: Results suggest that paclitaxel modifies gene and protein expression of skin-related molecular markers, and impairs different physical, physiological, and biomechanical properties of the skin of cancer patients at a subclinical level.Taxanes are chemotherapeutic agents that produce antitumor activity by causing stabilization of microtubules, thereby inhibiting cell cycle progression [1]. Paclitaxel (PTX) is the prototype of the taxane family of antitumor compounds and binds to the β-tubulin subunit in the microtubule, leading to its stabilization and increasing microtubule polymerization [2]. This unique mechanism of action differentiates paclitaxel from other antimicrotubule agents such as vinca alkaloids or colchicine, which inhibit tubulin polymerization. The microtubules formed in the presence of paclitaxel are so stable that they cause cell death by disrupting the normal microtubule dynamics required for cell division and interphase processes [3]. The consequent arrest of the cell cycle has been considered as the cause of paclitaxel-induced cytotoxicity. However, the signaling pathways that lead to apoptosis are not well understood. Recent discoveries indicate that paclitaxel initiates apoptosis through multiple mechanisms [4].In 1992, paclitaxel was approved by the US Food and Drug Administration (FDA) for the treatment of ovarian cancer. In 1996, a study of the Gynecologic Oncology Group (GOG) showed that paclitaxel-cisplatin was superior in terms of survival to the cyclophosphamide-cisplatin regimen as upfront therapy in stage III–IV ovarian cancer patients [5]. These results were confirmed by the European-Canadian Intergroup study [6]. These data justified the use of paclitaxel and platinum combination, and the treatment has become the standard of care in the first line. In the platinum-resistant setting, weekly paclitaxel has been considered one of the recommended regimens [7]. In endometrial and cervical cancer, paclitaxel in combination with platinum has also become part of the standard regimens in the first-line treatment. From this point, paclitaxel has been also used in the treatment of other cancers including colorectal and breast cancer, head and neck cancers, small-cell and non-small-cell lung cancers, and AIDS-related Kaposi Sarcoma [8].Although taxanes are tolerable and manageable, their toxic profile includes a wide number of adverse events. Hematological, cardiologic, and neurologic toxicities are very common in taxane-containing regimens [9]. Neutropenia is described amongst the principal toxic effects of PTX and is dose-limiting [10]. Peripheral neuropathy, another dose-dependent side effect, is found in 60–70% of chemotherapy patients and is characterized by sensory symptoms, such as numbness and paresthesia [11]. One of the most frequent adverse events is taxane-induced dermatologic toxicity, which has been reported in up to 89% of patients [12]. The spectrum of cutaneous reactions to paclitaxel includes alopecia, hypersensitivity reactions such as erythema and urticaria, nail changes, and radiation recall dermatitis. Less common effects such as acral erythema, erythema multiforme, pustular dermatitis, and scleroderma-like changes have also been described [12,13]. Generally, the adverse effects on the skin are mild to moderate in severity and self-limiting. Consequently, they are usually dose-dependent and sometimes require dose reductions, interruptions, or termination of the taxane chemotherapy [14].There are scarce data regarding the mechanisms that lead to these toxic effects and most remain not understood. Moreover, the real impact of the taxane-induced skin toxicity could be underestimated as the skin adverse events are usually under-reported or paucisymptomatic [14]. The available data in vivo are limited to case reports and oncology studies, and usually describe events happening to symptomatic patients. There is no information regarding the impact of taxanes on the skin in patients without cutaneous symptoms. However, a direct cytotoxic effect of chemotherapy on basal keratinocytes has been proposed; histology from skin biopsies of PTX treated patients have shown alterations in keratinocytes when some cutaneous events occur [15,16,17]. Further, studies in vitro have described that paclitaxel induces a cytotoxic response in transformed HaCat keratinocytes [18] and produces epithelial damage in zebrafish models [19,20]. Further, our previous results showed that paclitaxel impacts on the expression of proteins related to angiogenesis, elasticity, inflammation, and senescence in human keratinocytes [21].Of note, studies on undifferentiated keratinocyte monolayer cultures can lack some of the physiological functions of the stratified keratinocyte epithelium and could misinterpret the results obtained in preclinical studies. Thereby, various three-dimensional (3D) skin equivalents reproducing in vivo conditions have been developed for pharmacologic and toxicologic in vitro testing as an alternative to animal models [22,23]. One of these models is characterized by the growth of keratinocytes on a feeder layer of lethally irradiated 3T3 fibroblasts. The feeder layer supports and maintains keratinocyte colony growth and stratification [24,25], producing a 3D model that is compatible with autologous and allogenic transplantation [26,27].In this study, we aimed to overcome the lack of investigation regarding subclinical alterations induced by PTX. Therefore, we analyzed PTX-induced subclinical skin alterations by measuring different biomechanical properties of the skin in oncologic patients. Secondly, we reconstructed a 3D epidermis cell model to mimic a healthy epidermis and evaluate the effects of paclitaxel in some of the molecular markers associated with skin homeostasis. The results obtained will help understand asymptomatic skin alterations to prevent possible future skin adverse effects.This project was approved by the Research Ethics Committee of Valencia University Clinical Hospital and further authorized by the Valencian Regional Ministry of Health. Informed consent was obtained from each participant before starting the study. Twenty cancer patients and 20 healthy volunteers were recruited from the oncology service at Valencia University Clinical Hospital. The patients’ clinical features can be found in Table 1. Inclusion criteria for the control group comprised being Caucasian females aged 40–70 years old. Exclusion criteria included having an acute illness, skin pathologies, being pregnant, or breastfeeding. The inclusion criteria for cancer patients included: (1) being over 18 years old; (2) having a clinical diagnosis of gynecological cancer (ovarian, cervix or endometrium cancer at any stage of the International Federation of Gynecology and Obstetrics (FIGO) classification); (3) having indication for treatment with taxanes; (4) being treated for the first time or in relapse; (5) having adequate kidney, liver, and hematological functions before treatment; or (6) having PTX prescription on a 3 week schedule in combination or PTX in a weekly schedule either in monotherapy or in combination. Exclusion criteria comprised (1) having known chronic or rheumatologic skin disease; (2) being under corticosteroid treatment 2 weeks before admission to the study; (3) acute illness; (4) being pregnant or breastfeeding; or (5) having visible skin adverse effects. During the study, patients were not allowed to use cosmetic treatments.Skin parameter measurements were performed in three visits: before treatment (T1), during treatment after 3 chemotherapy cycles (T2), and at the end of treatment after 6 chemotherapy cycles (T3). The measurements were taken using specific probes following the measurement guidelines [28,29,30]. All probes were purchased from Courage–Khazaka Electronic (Cologne, Germany).Corneometer CM 825® (Courage-Khazaka Electronic, Cologne, Germany) was used to obtain cheekbone and forearm hydration values. This probe measures the electrical capacity of the stratum corneum, based on the linear dependency of the electrical property of the epidermis to its hydration. The results are displayed in arbitrary units [28].Tewameter TM 300® (Courage-Khazaka Electronic, Cologne, Germany) was used to provide cheekbone and forearm values of transepidermal water loss (TEWL). The probe measures the vapor pressure and calculates TEWL from the difference between two measurement points using Fick’s law of diffusion. It displays the results in grams per hour per square meter (g/hm2) [31].Sebumeter SM 815® (Courage-Khazaka Electronic, Cologne, Germany) was used to obtain the forehead sebum value. The probe determines the translucency of a special tape, which becomes transparent after contact with sebum on the skin surface and displays the sebum values in µg per square centimeter (µg/cm2) [32].Mexameter MX 18® (Courage-Khazaka Electronic, Cologne, Germany) was used to determine the forearm erythema based on tissue’s narrow wavelength light absorption. Results are displayed as erythema index in arbitrary units [29].Skin elasticity and firmness were assessed in the cheekbone with Cutometer® MPA 580. This probe suctions skin and gives a series of R values. The ones that provide more information about elasticity and firmness are R0, R2, R5, and R7. R0 expresses the maximum width of skin and is given in mm. R2 represents the ratio between the maximum width of the skin and its ability to return to its original state after suction (Ua/Uf). R5 represents the ratio between the elasticity of the suction phase and the elasticity of the relaxation phase (Ur/Ue). R7 represents the elastic recovery ratio (Ur/Uf). The closer R2, R5, and R7 are to 1, the greater the elasticity.Skin smoothness and roughness were assayed in the cheekbone by Visioscan® (Courage-Khazaka Electronic, Cologne, Germany). The device takes grayscale photographs to study the epidermis surface and analyzes the obtained images, where white and bright gray are associated with a bad condition of the skin. The software uses the images to determine the skin topography parameters SELS (surface evaluation of the living skin) [33]. The given parameters are smoothness (Sesm) and roughness (Ser) and are expressed in arbitrary units. Low values of Sesm inform about smoother skin, while high values of Ser imply rougher skin [34].Visioscan® was also combined with the Corneofix® (Courage-Khazaka Electronic, Cologne, Germany) technique to obtain the skin desquamation index on the forehead. Corneofix® is a tape that adheres to the skin and collects corneocytes. Then, the tape is placed on the Visioscan® probe which takes an image and processes it with a determined color grading, depending on the desquamation level. The color scale ranges from cool to warm colors. The software also analyzes the number, size, and thickness of the attached corneocytes to the Corneofix® tape and gives the desquamation percentage.Finally, Ultrascan® UC22(Courage-Khazaka Electronic, Cologne, Germany) was used to analyze the thickness of the different layers of the skin on the forearm. The probe takes an ultrasound image of the skin.All measurements were taken under controlled conditions. Temperature was maintained at 22 ± 2 °C and relative humidity between 40% and 60%. Measurements were taken after patients remained in a 30 min acclimatization period in the same atmospheric conditions.Three-dimensional epidermis cell models were reconstructed using the BALB/3T3 feeder-layer technique adapted from Mak et al. [35] and Arnette et al. [24]. In brief, 106 BALB/3T3 fibroblasts (Lonza, Basel, Switzerland) were seeded on collagen-coated Millicell inserts (Millicell-CM 12 mm, transparent Biophore Membrane; Millipore Corp., Bedford, MA, USA) and placed into 6-well plates (Corning Incorporated, Corning, NY, USA). Fibroblasts were cultured for 2 days in 1 mL Dulbecco’s Modified Eagle Medium (DMEM, high glucose; Gibco, Waltham, MA, USA) supplemented with 10% fetal calf serum (FCS, Gibco, Waltham, MA, USA) and added to the apical and dorsal side of the insert. When fibroblasts reached 60–70% confluence, the monolayer was irradiated with UV light at 0.048 mW for 1 h with UVACUBE 400 (Honle UV Technology, Gräfelfing, Germany) to establish the feeder layer. Then, primary adult epidermal keratinocytes (192627, Lonza, Basel, Switzerland) were seeded at a density of 0.5 × 106 cells/cm2. Cultures were grown at 37 °C and 95% air/5% CO2 until approximately 60% confluency and then were switched to Keratinocyte Growth Medium (KGM-Gold™, Lonza, Basel, Switzerland) supplemented with KGM-Gold SingleQuot Kit (Lonza) until confluent. Confluent cultures were raised to the air–liquid interface and cultured for 21 days until epidermal stratification was achieved. To validate the stratification, histological analysis was performed after 21 days. The reconstructed epidermis tissues were fixed with 10% formalin solution, dehydrated, and embedded in paraffin. Six-micrometer-thick sections were cut and stained with hematoxylin–eosin. Random photographs were taken of each sample with a Leica DM6000B microscope (Leica Biosystems, Wetzlar, Germany).The 3D epidermal cell models were incubated for 24 h with PTX within the clinically achievable plasma concentrations of 0.3, 3, and 30 µM [36,37,38]. After incubation, total RNA was extracted using TRIzol® Reagent (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) following the manufacturer’s instructions. Reverse transcription was performed in 500 ng of total RNA with a TaqMan reverse transcription reagents kit (Applied Biosystems, Thermo Fisher Scientific, Waltham, MA, USA). cDNA was amplified with specific primers and probes predesigned by Applied Biosystems for aquaporin 3 (AQP3) (Hs00185020_m1), collagen type 1 (COL1) (Hs00164004_m1), elastin (ELN) (Hs00355783_m1), and fibronectin (FN1) (Hs01549976_m1) in a QuantStudio™ 5 Real-Time PCR System, using universal master mix (Applied Biosystems, Thermo Fisher Scientific, Waltham, WA, USA). Expression of the target gene was expressed as the fold increase or decrease relative to the expression of β-actin (Hs01060665_g1) as an endogenous control. The mean value of the replicates for each sample was calculated and expressed as the cycle threshold (Ct). Gene expression level was calculated as the difference (ΔCt) between the Ct value of the target gene and the Ct value of β-actin. The fold changes in the target gene mRNA levels were designated 2−ΔCt.The 3D epidermal cell models were incubated for 24 h with different PTX concentrations (0.3, 3, and 30 µM). After incubation, protein extraction was performed incubating samples with lysis buffer (1M HEPES, 4 M NaCl, 0.5 M EDTA, 0.1 M EGTA) supplemented with a protease inhibitory cocktail complete™ and phenyl-methyl-sulfonyl fluoride (PMSF) (Roche Diagnostics, Indianapolis, IN, USA). Total protein concentration was quantified using the BCA Protein Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA). Protein electrophoresis was performed to separate proteins according to their molecular weight. Twelve micrograms of denatured proteins along with Rainbow ™ molecular weight marker (Sigma-Aldrich, Saint Louis, MO, USA) were loaded into Mini-PROTEAN® polyacrylamide gels TGX™ (Bio-Rad, Herts, UK) by application of 100 V during 1 h. Proteins were transferred from the gel to a nitrocellulose membrane Trans-Blot® Turbo™ Transfer Pack, using the Trans-Blot® Turbo™ Transfer System (Bio-Rad Laboratories, Herts, UK). Then, membranes were incubated with 5% bovine serum albumin (BSA) for 2 h and labeled overnight at 4 °C, with various primary antibodies. The secondary antibody was incubated for 1 h at room temperature. The primary antibodies used were the following: AQP3 (ab125219, Abcam, Cambridge, UK); COL1A (PA5-95137, Thermo Fisher Scientific, Waltham, MA, USA); FN1 (PA5-29578, Thermo Fisher Scientific, Waltham, MA, USA); and ELN (ab23747, Abcam, Cambridge, UK). To normalize results, the β-actin antibody (A1978, Sigma-Aldrich, Saint Louis, MO, USA) was used as housekeeping control. Signal visualization of proteins was carried out by incubating the membranes with chemiluminescence reagents (ECL Plus, Amersham GE Healthcare, Buckinghamshire, UK). Densitometry of films was performed using Image J 1.42q software. Results of target protein expression are expressed as the percentage of the densitometry of the endogenous controls β-actin.Results from cellular in vitro experiments were expressed as the mean ± standard error (SE) of n experiments; p < 0.05 was considered statistically significant. Normal distribution for each data set was confirmed by the Kolmogorov–Smirnov test. Statistical analysis was carried out by multiple comparisons with analysis of variance (ANOVA) followed by Bonferroni post hoc test. Results from the human in vivo experiments were expressed as the mean ± standard error (SE) of n experiments; p < 0.05 was considered statistically significant. When the comparisons concerned only 2 groups (healthy vs. PTX T1), statistical analysis was carried out by unpaired t-test. Multiple comparisons were analyzed by ANOVA followed by the Bonferroni post hoc test.The stratification of the 3D epidermis cell model was confirmed by the hematoxylin–eosin staining. As shown in Figure 1A, keratinocytes were distributed into the principal epidermis layers: basal, spinous, and granular, and its terminal differentiation resulted in the presence of the stratum corneum, analogously to the epidermal in vivo structure of healthy skin.The effects of PTX treatment in the hydration molecular marker aquaporin (AQP3) in the 3D skin model were examined. Incubation with PTX for 24 h induced a dose-dependent decrease in gene expression that was statistically significant (Figure 1B). Further, incubation of the 3D skin model with PTX for 24 h induced a similar significant decrease in AQP3 protein expression in all doses (Figure 1C). Hydration was also measured with Corneometer® on the cheekbone and forearm of the oncologic patients and to the control group (Figure 1D). No statistically significant differences were observed between the values of the control subjects and the values of cancer patients before treatment (T1). However, after three (T2) and six (T3) PTX cycles, both areas of skin manifested reduced hydration values. At the T3 timepoint, the variation percentages were −18.73% ± 5.42 and −16.38% ± 5.67 in the cheekbone and forearm, respectively. TEWL was also examined in the cheekbone and forearm with Tewameter® (Figure 1D). No statistically significant differences were observed in TEWL between the control group and the cancer patients before treatment (T1). Chemotherapeutic treatment with PTX decreased the TEWL value in the second (T2) and third (T3) visits. The differences were statistically significant in the forearm area at the T3 timepoint, and in the cheek area at T2 and T3 timepoints (Figure 1B). Medium TEWL percentage variations at T3 were −26.67% ± 8.63 in the cheekbone and −18.56% ± 9.49 in the forearm.Incubation of the 3D skin model with PTX for 24 h induced a decrease in the gene expression of the three analyzed skin elasticity and firmness markers: COL1, ELN, and FN1. The mRNA downregulation was dose-dependent (Figure 2A). In the same way, treatment with PTX for 24 h induced the same response at the protein level, decreasing COL1, ELN, and FN1 protein expression in all doses (Figure 2B).The effect of PTX on the elasticity and firmness of the skin was also evaluated in oncologic patients using Cutometer® (Figure 2C). In all parameters analyzed no differences were observed between the healthy volunteers and the oncologic patients before treatment (T1). However, treatment with PTX induced significant reductions in all values at T2 and T3 timepoints. The mean variations at T3 were R0: −47.21% ± 8.33, R2: −19.69% ± 6.10, R5: −33.06% ± 5.92, and R7: −19.40% ± 7.09.The effects of PTX on skin lipids were evaluated by measuring the skin sebum production with Sebumeter® (Figure 3A), and its effects on skin redness were evaluated by measuring the erythema value obtained with Mexameter® (Figure 3B). In both parameters, no differences were encountered between the healthy group and the oncologic group previously to treatment (T1). After three PTX cycles (T2), the sebum levels were slightly reduced but not significant, while the erythema values remained constant. After six PTX cycles (T3), sebum levels were reduced significantly with a variation percentage of −45.29% ± 8.23 and the erythema value increased with a mean variation of 13.96% ± 4.11.The smoothness (Sesm) and roughness (Ser) parameters were also evaluated on the skin of oncologic patients with the Visioscan® equipment. As seen on the representative images in Figure 4C, treatment with PTX worsened the aspect of the skin surface at T2 and T3 timepoints. The Visiosca® software analyzed each image and displayed the Ser (Figure 4A) and Sesm (Figure 4B) parameters. Ser is directly proportional to roughness and Sesm is inversely proportional to smoothness. No differences were observed between the Ser and Sesm values of healthy volunteers and oncological patients before treatment (T1). However, chemotherapeutic treatment with PTX produced an increase in Ser and Sesm values, after six cycles of PTX (T3) (Figure 4A,B), which represent a loss in skin smoothness and an increased roughness. The mean Ser variation at T3 was 49.02% ± 17.94 and the mean Sesm variation was 20.54% ± 5.80.The effect of PTX on skin desquamation was evaluated by combining the Corneofix® tape sheets with the Visioscan® equipment. Visioscan® software displays a color-graded image of the Corneofix® tape with the attached corneocytes. As shown in the representative images in Figure 5A, higher blue staining at T2 and T3 represent higher desquamation levels. The analyzed desquamation percentage is shown in Figure 5B. While no differences were observed between the percentage of desquamation of healthy volunteers and cancer patients at T1, treatment with PTX increased skin peeling, with statistically significant differences after six PTX cycles (T3). The mean variation at T3 compared to T1 was 14.94% ± 3.17.The ultrasound technique of Ultrascan® UC22 was used to analyze the thickness of the different layers of the skin in oncologic patients. Epidermis, dermis, and the total thickness of the skin were similar in the healthy group and patients before treatment. After three and six PTX cycles, all layers of skin showed a decrease in thickness, which can be noted in the representative images in Figure 6A. The mean thickness variations in comparison to T1 were −34.37% ± 3.50 in epidermis, −12.75% ± 2.05 in the dermis, and −16.46% ± 1.58 for total skin thickness (Figure 6B).Paclitaxel is an antineoplastic drug widely used in cancer treatment that has been shown to produce a multitude of skin adverse effects [39]. However, the subclinical alterations caused by PTX on the skin have not previously been described. In this study, we investigated these events in oncologic patients under PTX treatment by measuring hydration, TEWL, sebum, elasticity, erythema, roughness, desquamation, and thickness of the skin. The study was carried out without the development of skin reactions. The use of the Courage–Khazaka Electronic probes allowed us to measure the different skin parameters without the need of performing histological analysis of skin biopsies. To support these findings with in vitro experiments, we used a 3D epidermis model of keratinocytes grown on a feeder layer and exposed to the air–liquid interface. Hematoxylin–eosin staining demonstrated the development of a fully differentiated epidermis. The 3D epidermis model was used as a mimicker of a healthy epidermis to evaluate molecular modulation induced after treatment with the clinically achievable plasma concentrations of PTX.In this study, oncologic patients did not report visible cutaneous symptoms. The values of the skin properties in oncologic patients before treatment (T1) were compared to the values of a healthy control group. In all cases, the parameters were similar and no statistically significant differences were encountered between both groups. This shows that before treatment with PTX, all patients had skin parameters within the normal biological values, representative of the general population. Additionally, data obtained from healthy volunteers and cancer patients in their first visit were within the reference values, as described by the literature [29,30,40,41]. Firstly, we analyzed the hydration levels and TEWL. In the epidermis cell model, treatment with PTX reduced the gene and protein expression of the marker Aquaporin (AQP3). AQP3 is the most abundant skin aquaglyceroporin and is responsible for transporting water and glycerol in the epidermis. Therefore, AQP3 is a key protein in the maintenance of epidermal hydration and differentiation of keratinocytes [42,43,44]. Only one case report has related PTX and aquaporins previously: A patient under PTX treatment developed a cystoid macular edema induced by the functional failure of aquaporin mediated water transport [45]. This finding, in line with our results, suggests that PTX induces the modulation of aquaporins. In addition, aquaporins are reported to play a major role in angiogenesis, cell proliferation, apoptosis, and cell migration [46]. Thereby, PTX could be impairing other cellular processes through the modulation of AQP3. To see if these molecular changes could directly affect the skin in patients under PTX treatment, we measured hydration levels and transepidermal water loss (TEWL). In agreement with the in vitro results, patients showed reduced face hydration levels after three and six PTX cycles. Hydration values on the forearm were significantly reduced after six PTX cycles. This difference can be explained by the fact that the face is a photoexposed area, and the damage produced by sunlight on skin cells might enhance the dehydration induced by PTX [47,48]. Skin dehydration in oncologic patients can lead to the reduction of the skin water content. Indeed, patients showed a decrease in TEWL in comparison to the values obtained before treatment. TEWL is an indicator of the ability of the epidermis to hold water and is a good marker of the functionality of the skin barrier [30]. In patients with taxane-related xerosis, TEWL appears increased [14]. We suggest that these differences might be explained by the fact that patients with xerosis have a developed skin adverse effect, while patients in this study did not show any symptomatic alterations. The effects of PTX in this study represent the early modifications that could lead to the development of later adverse effects. Furthermore, because lipids act as a barrier against water loss [49], we also analyzed the sebum levels in oncologic patients. Sebum levels showed a progressive decrease with the progression of PTX treatment, which represents the loss of its skin protection layer. Overall, we can state that PTX proved to impair skin moisturization in oncologic patients and compromised the skin barrier function.The epidermis cell model was also incubated with PTX to analyze collagen 1 (COL1), elastin (ELN), and fibronectin (FN1) expression. These proteins play a key role in maintaining the elasticity, firmness, and support of the skin layers [50,51,52]. Gene and protein expression of these markers was significantly reduced after PTX treatment on the 3D epidermis model. In agreement with these results, a study on tenon fibroblasts monolayers showed that both collagen and fibronectin were markedly downregulated in the culture medium [53,54]. There is also in vivo evidence that showed alterations in collagen in a skin biopsy from a sclerodermatous area of a patient under taxane treatment [55]. These results are directly related to those obtained in our in vivo study on cancer patients treated with PTX. Oncologic patients had decreased skin elasticity and firmness as shown by the lowered R parameters, which represent the state of the biomechanical elastic properties of the skin [56,57,58]. These results suggest that PTX-induced decrease in the skin elasticity and firmness might be mediated by its capacity to modulate molecular markers such as COL1, ELN, and FN1, which maintain the structure of the skin layers.Erythema is a common adverse effect seen on patients under PTX treatment [13,59]. In this study, patients did not develop clinically visible erythema. However, we wanted to analyze the possible subclinical manifestations. Colorimetric examinations with Mexameter® showed that the erythema index had a slight increase after six PTX cycles. It is widely described that treatment with taxanes induces erythema at different areas of skin and in different grades of severity [14,60,61]. However, this is the first study that reported erythemal changes on the skin under PTX treatment before the development of clinically visible erythematous skin reactions. As it was evident that PTX induced changes in the biomechanical properties of skin, roughness and softness were also examined. PTX induced an increase in skin roughness and a decrease in smoothness. Connected to these events, patients showed an increased desquamation percentage. This implies a decrease in the barrier function capacity of the skin after PTX treatment. Similar to the other parameters analyzed, the research data available comes from case reports of patients suffering from PTX-induced toxicities with clinical manifestations. In this case, desquamation had been previously described in patients under PTX treatment, in the palms and soles, necessitating treatment interruption [62] and associated with rash reactions [63]. Regarding the thickness of the skin, PTX induced the reduction of the dermis, epidermis, and total skin thickness. These results were expected, bearing in mind that PTX induced the reduction of the structural proteins COL1, FN1, and ELN, which causes an impairment of the skin’s architecture, and therefore changes in its thickness. This effect of PTX has not been described previously. However, there is a case report describing acanthotic epidermis in a PTX-induced cutaneous eruption [16], which indicates the potential of PTX to induce changes in the structure of the skin.There are a few study limitations to be mentioned. Firstly, the heterogeneity of the sample—there are patients with localized disease and patients in different situations of advanced disease. Therefore, the chemotherapy administration schemes differ between patients. However, the heterogeneity of the sample should not have a significant impact since it includes only gynecological tumors with similar treatments. Secondly, considering the treatments used, the drug with the greatest impact at the cutaneous level was PTX. We acknowledge that carboplatin causes skin adverse effects but they are usually associated with hypersensitivity reactions [64]. As the number of patients was limited, and the purpose of the study was to analyze asymptomatic skin alterations, it was considered that the other drugs used in combination with PTX will not significantly bias our data. Finally, as not all patients were exposed to taxanes for the first time, a healthy population was included as a control group to prove that the baseline results of cancer patients before treatment were not statistically different from the healthy volunteers. This comparison served to prove that, in the patients included in the study, accumulated doses of paclitaxel did not affect the severity of the alterations.The results of this study show that, although PTX did not cause severe skin adverse reactions, it impaired its physical, physiological, and biomechanical properties with no clinical manifestations. Treatment with PTX induced skin dehydration, a decrease in elasticity, thickness, and sebum levels, and an increase in skin desquamation and erythema. These changes were related to the modulation of gene and protein expression induced by PTX in an epidermis cell model that mimics a healthy epidermis. These results indicate that PTX can alter the skin structure, and impair its barrier function, inducing cutaneous changes that do not become symptomatic. Previous studies have described PTX-induced epithelial damage in zebrafish models [19,20] and some case reports show histologic changes caused by PTX on the skin [16,17]. However, to our knowledge, this is the first study that analyzes the subclinical alterations caused by PTX and it might explain the prediction of later severe cutaneous adverse events. Skin symptoms in patients cause physical pain and discomfort and psychological distress. In severe cases, skin toxicities can cause treatment delays and even discontinuation, which affects clinical outcomes [65]. This highlights the need for the early management of these alterations. There is a relative lack of evidence for effective management of taxane skin toxicities [66]. Some studies recommend using a scalp cooling system to reduce alopecia, frozen gloves to prevent nail and cutaneous hand changes [12], and nail solution to prevent chemotherapy-induced nail toxicity [67]. Additionally, a preliminary study in patients undergoing chemotherapy with taxanes demonstrated that resveratrol, lycopene, vitamin C, and anthocyanins (Ixor®) had a protective role against skin reactions [68]. However, there is still a lack of studies that propose prophylactic measures to prevent skin alterations in patients under paclitaxel treatment.In conclusion, the results provided by this work suggest the need for prophylactic measures that improve the patient’s quality of life as well to ensure adherence to treatment. Although more studies addressing this matter are needed, early introduction of effective countermeasures including daily skincare (skin cleansing, moisturization, and irritation prevention) would help in the prevention of future PTX-induced skin toxicities.This work has shown that paclitaxel impairs different physical, physiological, and biomechanical properties of the skin. To our knowledge, this is the first study that has concluded that gynecological cancer patients under paclitaxel treatment show subclinical skin alterations. These subclinical alterations include the decrease of hydration, TEWL, sebum, elasticity, and thickness of the skin, together with an increase in erythema, roughness, and desquamation. Further, this study showed the PTX-induced modulation of molecular markers related to hydration and the support of the skin layers in a 3D epidermis model. These altogether highlight the lack of management measures to prevent skin alterations in patients under taxane treatment. The use of prophylactic measures at the early stages of treatment could be useful to avoid these subclinical alterations and prevent future severe reactions.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051146/s1, Supplementary file: un-cropped images of the original western blots from which figures have been derived.Conceptualization, J.A.P.-F., C.S., A.C., and J.C.; methodology P.M., M.P.-L., C.E., and I.R.; validation, J.A.P.-F., C.S., J.M., A.C., and J.C.; formal analysis, P.M., M.P.-L., C.E., and I.R.; investigation, P.M., M.P.-L., C.E., and I.R.; resources, J.C.; writing—original draft preparation, P.M.; writing—review and editing, P.M., M.P.-L., and J.C.; supervision, J.C.; project administration, J.C.; funding acquisition, J.C. All authors have read and agreed to the published version of the manuscript. This work was supported by the grants PID2020-114871RB-I00 (JC), European Regional Development Fund (FEDER), and Instituto de Salud Carlos III, PI20/01363 (JM), CIBERES (CB06/06/0027) from the Spanish Government, and by research grants from the Regional Government Prometeo 2017/023/UV (JC), from “Generalitat Valenciana”. Funding entities did not contribute to the study design or data collection, analysis, and interpretation, or to the writing of the manuscript.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the research Ethics Committee from the Clinical Hospital from Valencia (approval date 29 November 2018) and by the Council of Public Health from Generalitat Valenciana (date of approval 2 February 2019), ethical approval code: INC-PAC-2018-01.Informed consent was obtained from all subjects involved in the study. Written informed consent has been obtained from the patient(s) to publish this paper.The data presented in this study are available on request from the corresponding author.The authors declare no conflict of interest.Paclitaxel induces a dose-dependent loss of hydration levels in oncologic patients and reduces the expression of the hydration marker AQP3 in the 3D epidermal cell model. (A) Paraffin section from the 3D epidermis model stained with hematoxylin and eosin. Scale bar 100 µm. (B) Three-dimensional epidermal cell model tissues were incubated for 24 h with increasing paclitaxel (PTX) concentrations. Aquaporin (AQP3) mRNA levels were measured by real-time PCR. Data are expressed as 2−ΔCt. (C) Three-dimensional epidermal cell model tissues were incubated for 24 h with increasing paclitaxel (PTX) concentrations. AQP3 protein levels were analyzed by Western blotting. Quantification was performed by densitometry and normalized to β-actin. Results are expressed as the mean ± standard deviation of two independent experiments (n = 3); * p < 0.05 vs. control. Uncropped Western Blots can be found at Supplementary File. (D) Hydration levels were measured in 20 oncologic patients before (T1), during (T2), and after (T3) treatment with PTX, and in 20 healthy subjects as a control group. Measurements were conducted with Corneometer CM 825® in the cheekbone and forearm. Transepidermal water loss (TEWL) levels were measured in 20 oncologic patients before (T1), during (T2), and after treatment (T3) with PTX, and in 20 healthy subjects as a control group. Measurements were conducted with Tewameter TM 300® in the cheekbone and forearm. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1.Paclitaxel impairs skin elasticity and firmness in oncologic patients and reduces the expression of elasticity and firmness molecular markers in a 3D epidermis model. (A) The 3D epidermis models were incubated for 24 h with increasing paclitaxel (PTX) concentrations. Collagen type 1 (COL1), elastin (ELN), and fibronectin (FN1) mRNA levels were measured by real-time PCR. Data are expressed as 2−ΔCt. (B) The 3D epidermis models were incubated for 24 h with increasing PTX concentrations. COL1, ELN, and FN1 protein levels were analyzed by Western blotting. Quantification was performed by densitometry and normalized to β-actin. Results are expressed as the mean ± standard deviation of two independent experiments (n = 3); * p < 0.05 vs. control. Uncropped Western Blots can be found at Supplementary File. (C) The elasticity and firmness parameters R9, R2, R5, and R7 were measured in 20 oncologic patients before (T1), during (T2), and after (T3) treatment with PTX, and in 20 healthy subjects as a control group. Measurements were conducted with Cutometer® MPA 580 probe in the cheekbone. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1.Paclitaxel reduces sebum levels and increases erythema in oncologic patients. (A) Sebum levels were measured in 20 oncologic patients before (T1), during (T2), and after (T3) treatment with PTX, and in 20 healthy subjects as a control group. Sebumeter SM 815® was used to obtain the forehead sebum value. (B) The erythema index was measured in 20 oncologic patients before (T1), during (T2), and after treatment (T3) with PTX, and in 20 healthy subjects as a control group. Mexameter MX 18® was used to determine the forearm erythema. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1.Paclitaxel induces an increase of skin roughness and reduces its smoothness in cancer patients. (A,B) Skin roughness and smoothness were measured in 20 oncologic patients before (T1), during (T2), and after (T3) treatment with paclitaxel (PTX), and in 20 healthy subjects as a control group. Measurements were conducted in the cheekbone by Visioscan® VC 98 probe, obtaining the Ser and Sesm values calculated by the software from each grayscale photograph. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1. (C) Representative images of the skin topography obtained with Visioscan® during PTX treatment at time points T1, T2, and T3. White and bright gray on the images are associated with a bad condition of the skin. Scale bar 1 mm.Paclitaxel increases the skin desquamation levels in oncologic patients. (A) Representative Corneofix® tape images obtained before (T1), during (T2), and after (T3) treatment with paclitaxel. Images were obtained with the Visioscan® probe. The blue color on the image represents higher desquamation levels. Scale bar 1 mm. (B) Skin desquamation percentage was measured in 20 oncologic patients after PTX treatment at T1, T2, and T3 timepoints and in 20 healthy subjects as a control group. Measurements were conducted on the forehead with the combination of the Corneofix® tape and the Visioscan® probe. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1.Paclitaxel reduces skin thickness in oncologic patients. (A) Representative ultrasound images of the skin thickness obtained during treatment with PTX in 20 oncologic patients before (T1), during (T2), and after (T3) treatment. Images were obtained with Ultrascan® UC22. Scale bar 1 mm. (B) Epidermis, dermis, and total skin thickness measured in 20 oncologic patients under PTX treatment at timepoints T1, T2, and T3, and in 20 healthy subjects as a control group. Measurements were conducted on the forearm with Ultrascan® UC22. Results are expressed as the mean ± standard deviation of at least 3 measurements each time (n = 20); * p < 0.05 vs. T1.Patients’ clinical features.Grade: Depends on the cancer aggressiveness—for endometrial and endometrioid-type ovarian tumors, grades are classified according to the classic system from 1 to 3. For ovarian tumors, the current high-grade versus low-grade grading system is shown. Stage: International Federation of Gynecology and Obstetrics (FIGO) classification scale from I to IV. Strategy: For endometrial and cervical tumors, it is reported whether it was for localized disease (adjuvant) or the number of chemotherapy lines for advanced disease (1st line or 2nd line). For ovarian tumors, it is reported if the strategy was for localized disease (adjuvant or neoadjuvant) or advanced disease according to previous platinum sensitivity (platinum-sensitive or platinum-resistant). Refractory cancers were categorized as platinum-resistant. Treatment: weekly paclitaxel (80 mg/m2) was administered at days 1, 8, and 15 every 21 days. Three weekly paclitaxel (175 mg/m2) + carboplatin (AUC5) was administered in day 1 every 21 days. Abbreviation: NS: not specified; Pt-sensitive: platinum-sensitive; Pt-resistant: platinum-resistant.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01147.txt

@@ -0,0 +1 @@
+In oncology, treatment outcomes can be competing, which means that one treatment could benefit one outcome, like survival, and negatively influence another, like independence. The choice of treatment therefore depends on the patient’s preference for outcomes, which needs to be assessed explicitly. Especially in older patients, patient preferences are important. Our systematic review summarizes all studies that assessed patient preferences for various treatment outcome categories. A total of 28 studies with 4374 patients were included, of which only six studies included mostly older patients. Although quality of life was only included in half of the studies, overall quality of life (79%) was most frequently prioritized as highest or second highest, followed by overall survival (67%), progression- and disease-free survival (56%), absence of severe or persistent treatment side effects (54%), treatment response (50%), and absence of transient short-term side effects (16%). In shared decision-making, these results can be used by healthcare professionals to better tailor the information provision and treatment recommendations to the individual patient.For physicians, it is important to know which treatment outcomes are prioritized overall by older patients with cancer, since this will help them to tailor the amount of information and treatment recommendations. Older patients might prioritize other outcomes than younger patients. Our objective is to summarize which outcomes matter most to older patients with cancer. A systematic review was conducted, in which we searched Embase and Medline on 22 December 2020. Studies were eligible if they reported some form of prioritization of outcome categories relative to each other in patients with all types of cancer and if they included at least three outcome categories. Subsequently, for each study, the highest or second-highest outcome category was identified and presented in relation to the number of studies that included that outcome category. An adapted Newcastle–Ottawa Scale was used to assess the risk of bias. In total, 4374 patients were asked for their priorities in 28 studies that were included. Only six of these studies had a population with a median age above 70. Of all the studies, 79% identified quality of life as the highest or second-highest priority, followed by overall survival (67%), progression- and disease-free survival (56%), absence of severe or persistent treatment side effects (54%), and treatment response (50%). Absence of transient short-term side effects was prioritized in 16%. The studies were heterogeneous considering age, cancer type, and treatment settings. Overall, quality of life, overall survival, progression- and disease-free survival, and severe and persistent side effects of treatment are the outcomes that receive the highest priority on a group level when patients with cancer need to make trade-offs in oncologic treatment decisions.Being diagnosed with cancer is a major life event and the start of a complex decision-making process on cancer treatment. Often, several treatment options are available. Most cancer treatments are intensive and burdensome, and the outcome cannot be guaranteed [1,2]. Furthermore, outcomes can be competing. For example, adjuvant chemotherapy in stage III colon cancer may decrease the likelihood of recurrence and increase (cancer-specific) survival, but toxicity may impact quality of life in the short term while serious treatment-related complications could also impact long-term functioning. Trade-offs are therefore needed.The gold standard for complex decisions in oncology is shared decision-making, of which an important step is explicitly discussing which outcomes matter most to the patient [3,4,5,6]. This is particularly relevant in older patients, who may have a less favorable balance of benefits and risks of treatment than younger patients [1,7,8,9]. They are often excluded from clinical trials, and as a consequence, their recommendations are less evidence based [10]. Furthermore, oncological treatments have a narrow therapeutic index between the possible benefit of cancer control, including cancer symptom reduction, and the price that is still considered acceptable in terms of side effects. This increases the uncertainty in decision-making and makes it even more important to know which treatment outcomes are most frequently prioritized by older patients with cancer.Knowledge of the most frequently mentioned patient priorities allows for a tailoring of information provision and prevents information overload caused by summing up all the treatment and outcome possibilities during the shared decision-making [3,4,5,6]. Prior research has demonstrated that adequate information provision about treatment impact and adverse events reduces the likelihood of decision regret [11] and improves patient satisfaction [12].In patient preference elicitation many methods exist. Some methods are more general and ask patients to explicitly indicate what they would prefer, like rating scales [13] or the Outcome Prioritization Tool (OPT), which explicitly asks patients to rate each outcome relative to other outcomes without having two values on the same level [14]. This uses a trade-off principle: By prioritizing one outcome, patients are willing to accept the deterioration of other outcomes. The outcomes that are assigned priorities in the OPT conversation include extending life, maintaining independence, reducing pain, and reducing other symptoms [15]. Other methods are more specific and implicit, like discrete choice experiment (DCE), conjoint analysis (CA), and probability trade-off (Trade-off). These methods present patients with hypothetical scenarios with information on the possible benefits and side effects that are associated with various treatments and the probability of those happening. By measuring the willingness of patients to choose a treatment option while providing different scenarios of the included variables, the relative importance of that variable can be calculated [13]. Furthermore, the analytic hierarchy process (AHP) gives patients pair-wise comparisons and asks them to rate them against each other. This is also leads to a calculated relative importance of all included variables [16]. All methods have their benefits, and the best method of preference assessment depends on the question it needs to answer and the (number of) trade-offs that are at stake [13].Two types of treatment outcomes are described in the literature [17,18]: disease-centered outcomes, which measure the objective effect of the treatment on the tumor and the adverse events, such as treatment response, toxicity, and disease-free survival, and patient-centered outcomes, which focus on the patient’s perception of health, quality of life, and functional outcomes like maintaining independence. To get a complete overview of the patient priorities in older patients with cancer, we set out to gather all available evidence from trade-off studies regarding treatment outcomes (both disease-centered outcomes and patient-centered outcomes).We performed a systematic review to collect all available quantitative evidence comparing the relative importance patients allocate to various patient- and disease-centered outcomes after a cancer diagnosis. During the process the PRISMA guidelines were used [19]. We registered the systematic review in the OSF registry from the Center for Open Science [20].On 22 December 2020, we performed a search in Embase and Medline with the following terms and their synonyms: “health or treatment outcomes,” “priorities,” “trade-offs,” and “cancer.” The full electronic search strategy is shown in Appendix A. The search was limited to studies on humans written in English and published in the past 15 years. After an initial search in older patients, which resulted in few specific data, the search was expanded to all ages.The titles and abstracts of all studies retrieved by the searches were assessed by one reviewer (N.S.) to determine which ones warranted further examination. All potentially relevant titles were subsequently screened independently as full text by two reviewers (N.S., A.W.). If no full text was found, the reviewers tried to find the final report of the study by using names of the different authors in combination with key words from the title. If none were found, the studies were excluded.We included original publications on the comparison of outcome priorities after a diagnosis of cancer; this included both studies in actual cancer patients and studies performed on other subjects asked to state their priority in the hypothetical situation of a cancer diagnosis. Studies were only included if they addressed at least three of the possible six outcome categories that were defined, which were transient short-term side effects, severe and persistent side effects, quality of life (including functioning), treatment response, progression- and disease-free survival, and overall survival (see Appendix B).All methods of preference elicitation were allowed, as long as the studies provided a form of prioritization of the individual outcome categories. Therefore, studies were excluded if the relative importance of outcome categories could not be elucidated due to the way the results were elicited or reported.Both reviewers (N.S., A.W.) independently extracted the following characteristics: title, author, year of publication, country, cancer type, curative or palliative treatment setting, me(di)an age of the sample, sample size, and method of assessing preferences. In addition, any patient- or disease-centered outcomes that were included in the trade-offs in the study were extracted, together with the ranking or score regarding the priority for each of these outcomes. Outcomes relating to process attributes such as mode of administration, frequency of administration, or out of pockets costs were not included.Quality assessment was carried out by two independent reviewers using a quality assessment based on the Newcastle–Ottawa Scale ([21]; N.S. and A.W.), adjusted for this purpose based on a validated checklist for conjoint analysis (Appendix C) [22,23,24]. Disagreements were discussed in a consensus meeting and in case of continuing disagreement, a third reviewer (M.H.) was consulted.Based on the outcomes used by the included studies, two reviewers (N.S., A.W.) defined six outcome categories: quality of life (including functioning), overall survival, progression- and disease-free survival, severe and persistent side effects of treatment, treatment response, and transient short-term side effects. Detailed definitions can be found in Appendix B.Using this classification, each assessed outcome was allocated to one of the defined outcome categories by two independent reviewers (N.S. and A.W.). The scores that the study reported were used to prioritize outcome categories to identify the highest and second-highest priority. The results were reported using descriptive data, describing the proportion of studies that prioritized each outcome category as highest or second highest in relation to the number of studies addressing that outcome.If multiple outcomes in the study were allocated to the same outcome category (e.g., diarrhea and nausea were both transient short-term side effects), an average score of these outcomes was used to decide on the prioritization order of the outcome categories. In case of discrepancies, items were discussed until consensus was achieved; if needed, a third reviewer (M.H.) was consulted. When a study assessed preferred outcomes with multiple methods, resulting in different prioritizations, only the discrete choice elicitation was used to limit the heterogeneity. To determine the robustness of the results, subgroup analyses were conducted for curative and palliative settings and for older patients (study populations with a median age of 70 years or higher).The search resulted in 7321 hits (2042 from Medline and 5279 from Embase). After removing 2072 duplicates and 5222 studies for other reasons (Figure 1), a total of 27 publications were included. Cross-referencing yielded one more publication, resulting in a total of 28 studies in this systematic review.The characteristics of 28 selected studies are summarized in Table 1. Most were published in the past five years. The total study population consisted of 4374 patients; the median sample size was 133 patients (range 36–419). The me(di)an age of participants varied between 35 and 78 years and six studies had a population with a median age over 70 years. The most frequently studied cancer type was gastrointestinal cancer (n = 8), followed by six studies that assessed various cancer types (see Table 1). Eight studies examined curative treatment, 11 studied palliative treatment, and nine studied both. The studies that assessed both often had a mix of all stages of cancer together. Various methods of preference elicitation were used. The majority of the studies (n = 15) used discrete choice elicitation, followed by conjoint analysis (n = 5), the Outcome Prioritization Tool (n = 3), and various types of rating scales. Both probability trade-off and the analytic hierarchy process were used in one study (Table 2).Figure 2 provides an overview of the quality assessment. Details of each study can be found in Appendix D. In general, the representativeness of patients was good, although a few studies asked patients to provide answers for a hypothetical situation—for example, what they would choose if they had a different type or stage of cancer [29,40]. Some studies did not clearly report how specific outcomes were selected [28,29], or did not describe selection procedures at all [41,47,48,50]. Additionally, sometimes it was unclear how quality of life or other attributes were defined or were described to patients [27,39]. The analysis and outcome reporting were heterogeneous, but overall well described.The 28 publications reported 30 prioritizations: Two studies had a separate prioritization for patients with curative and palliative stages of disease. The median amount of outcome categories per study was four (range 3–6). The most frequently assessed outcome category was severe and persistent side effects (24 studies, 83%), followed by overall survival and transient short-term side effects (both, n = 19, 66%). Quality of life and progression- and disease-free survival were assessed in 14 (48%) and 15 studies (52%), respectively. Only one study included all six outcome categories (28), and there was no outcome category that was assessed in all studies.For each study, the highest and second-highest outcome categories are identified and shown in Figure 3 and Table 2 relative to the number of studies that assessed that outcome category. For example, quality of life was assessed in total in 14 studies and was in 11 studies the highest or second-highest priority (n = 11/14, 79%). Overall survival (67%), progression- and disease-free survival (56%), and severe and persistent side effects (54%) were also commonly prioritized. When focusing only on studies addressing a palliative setting, quality of life and overall survival were most important (both in 75%); in contrast, progression- and disease-free survival (67%) and treatment response (67%) were given the highest priority in a curative treatment setting (Table 2, Figure 4).The higher the percentage, the more frequently that outcome category was prioritized. Top 1 and top 2 priorities are shown. Percentages are given relative to the number of studies that assessed that outcome category. In total, 30 rankings from 28 studies are included. n represents the number of rankings that assessed the category.In a subgroup analysis of six studies focusing specifically on older patients [28,36,37,41,47,50] (median age of the study population of 70 years or higher or separate data of this subgroup), quality of life and overall survival were included in five of the six studies. They were also the highest or second-highest priority in most of them (n = 5/6, 83%, and n = 4/6, 67%, respectively), followed by progression- and disease-free survival (n = ½; 50%), severe and persistent side effects (n = 1/3; 33%), treatment response (n = 1/5; 20%), and transient short-term side effects (n = 0/2; 0%; see Table 2, Figure 4).In this systematic review, we examined which outcomes of treatment matter most to patients with cancer. While we were particularly interested in the priorities of older patients, the search was expanded because very few studies exist on patient preferences in older adults with cancer. In the 28 included studies, quality of life received high priority most frequently, followed by overall survival, progression- and disease-free survival, and severe and persistent side effects. In the palliative setting and in the subgroup analysis of older patients, quality of life and overall survival were most often prioritized, but in the curative setting, progression- and disease-free survival and treatment response were more important. This suggests that even though quality of life is most important overall on a group level, priorities might change depending on the intent of the treatment, contextual factors, and the age of the patient.In this systematic review, in which the majority of the studies included younger patients, quality of life was often considered important. Although severe and persistent side effects were assessed in the majority of studies, quality of life was underappreciated and included only in half of the studies. These findings are in line with previous research [52,53,54,55,56,57,58]. In our subgroup analysis of older patients, quality of life was assessed in the majority (75%), and together with overall survival, most frequently prioritized (83%). The outcome category of quality of life included functional and other patient-centered outcomes, but often was described in an unspecific way. However, especially in older patients, specific components of quality of life like cognition and functional abilities are considered important and few patients are willing to trade cognition for survival [59,60]. Thus, due to the inclusion of all ages and the unclear descriptions of quality of life, our review might underestimate the importance of certain aspects of quality of life in older patients.A recent systematic review on information needs in older patients with cancer showed that after patients received a cancer diagnosis, their focus was on short-term issues like understanding the situation, treatment options, and other practicalities, whereas information on functioning, quality of life, and dealing with late effects were given lower priority [6]. However, decision regret is more often linked to negative long-term outcomes [61], something that was discussed in half of the patients [62]. Since our study also shows that both quality of life (79%) and severe and persistent side effects (54%) were more frequently prioritized outcomes than transient short-term side effects (16%), patients should be aided in assessing and explicitly expressing which long-term outcomes matter most to them.Although both patients and physicians consider efficacy and physical side effects important in treatment choice, patients also incorporate other factors in their decision-making, like the impact on their daily life, family responsibilities, and the ability to attend important life events [63]. In the translation of general treatment outcomes to their personal situation, patients might interpret the outcomes differently than the physician. In addition, they might not realize that one treatment might have multiple competing effects. If the patient’s interpretation is left unrevealed, this may lead to a treatment choice that may not provide the patient with the benefit that they desire, or may also lead to a negative effect that lessens the benefit. For example, a patient with metastatic cancer may state that extending survival is most important, with an unexpressed underlying desire to care for an ailing partner for as long as possible. If the physician then tailors the treatment to value extending life, intensive palliative chemotherapy may be started. Although this may increase survival, it may in fact also hamper the patient’s caregiving abilities due to the side effects of treatment.These unique patient values that underlie the preference are not easily incorporated into disease- and treatment-specific decision aids. Moreover, physicians are not good at estimating their patients’ preferences [50]. Thus it can be helpful to have an additional preference assessment conversation as part of decision-making. This will help to clarify what a priority of a specific outcome means to the patient and why it is important to them, and will prevent treatment selection based on wrong interpretations from the patient or misunderstandings by the physician.In our review, some studies [41,50] used a non-disease and non-treatment-specific generic communication tool developed for patients with multimorbidity by Terri Fried: the Outcome Prioritization Tool (OPT) [15]. During this conversation, the healthcare professional verifies whether he or she understands the trade-offs correctly and invites the patient to explain why the outcomes are important and how they were interpreted [14]. Although this tool has been used in oncology patients before [41,50,64], it might be worthwhile to adapt this tool specifically for cancer patients and the treatment decisions they have to make.This study has some limitations. Firstly, the studies were heterogeneous; various types of cancer and various tumor stages were included. Furthermore, the studies assessed the priorities with their own defined benefits and risks. Depending on what was most appropriate given the characteristics of cancer-specific treatment regimens, they each asked the outcome categories differently and used different levels of the various attributes. Although the studies had sufficient common denominators to allow for the categorization and combination of results, this does make comparison between the palliative and curative treatment settings more difficult. For example, in a curative setting, overall survival might not be prioritized when described as a small increase in an already high survival rate. However, in the situation of a metastatic disease with a poor prognosis, overall survival might be prioritized, because living a few more months might be important for a patient who is awaiting their first grandchild.Moreover, multiple methods of assessing these outcome preferences were used, all with their own benefits and risks of bias [13,65]. To be able to compare the various outcome categories relative to each other and to minimize the effect of chance on ending up as a high priority, only studies that assessed at least three categories were included. Studies comparing only two categories and studies where it was not possible to elucidate the relevance of the various outcome categories were excluded. This might have changed the results, but made comparisons possible and allowed an actual trade-off between the benefits and negative effects of treatments to be registered.Finally, due to the heterogeneity of methods and reporting, we had to simplify the results of the studies to a ranking. This does not give details on whether the outcome priorities are close together or far apart within a study or on how much benefit gain or risk avoidance leads to a change in priority; thus, some information may have been lost. It did, however, allow us to identify the most important outcome categories on a group level. In clinical practice, these could be presented to individual patients who need to make and define their own trade-offs anyway. This pre-selection may prevent them from being overwhelmed by too many choices.In conclusion, understanding how patients prioritize potential outcomes of oncologic treatment and the trade-offs they are willing to make is an important component of shared decision-making. Our systematic review shows that quality of life, overall survival, progression- and disease-free survival, and avoiding severe and persistent side effects of treatment are the outcomes that receive the highest priority in patients with cancer.Conceptualization, M.E.H.; methodology M.E.H., P.A.L.S. and A.W.; software, P.A.L.S.; validation, P.A.L.S., A.W. and M.E.H.; formal analysis, P.A.L.S. and A.W.; investigation, P.A.L.S. and M.E.H.; resources M.E.H.; data curation, P.A.L.S. and A.W.; writing—original draft preparation, P.A.L.S.; writing—review and editing, P.A.L.S., M.E.H., J.E.A.P., S.F., M.E.S., P.S., S.R. and S.O.; visualization, P.A.L.S. and M.E.H.; supervision, M.E.H. and J.E.A.P.; project administration, P.A.L.S.; funding acquisition, M.E.H. and P.S. All authors have read and agreed to the published version of the manuscript.This research was funded by GERONTE. The GERONTE project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 945218.Not applicable.Not applicable.The data presented in this study are available on request from the corresponding author.The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.The following search was performed on 22 December 2020, in both Embase and Medline (((“treatment*”[tiab] OR “health”[tiab]) AND “outcome*”[tiab]) OR “scenario*”[tiab] OR “vignet*”[tiab] OR (“research”[tiab] AND “agenda”[tiab])) AND ((“priorit*”[tiab] OR “preference*”[tiab] OR “trade”[tiab] OR “trade off*”[tiab]) AND (“patient”[tiab] OR “patients”[tiab]) AND (“cancer”[tiab] OR “oncolog*”[tiab] OR “malignan*”[tiab]). Limitations: humans, English or Dutch, 2006, and onward study selection.Descriptions of the Outcome Categories.Quality assessment based on the Newcastle–Ottawa Scale.Quality Assessment Per Study.Study selection.Quality assessment—risk of bias.Outcome category prioritization of all studies.Older patient subgroup analysis. Outcome category prioritization of all studies (n = 28) compared to the subgroup of older patients (n = 6). A higher percentage means more frequently prioritized as the first or second priority. Percentages are relative to the number of studies that included that outcome category.Included studies and their characteristics.NSCLC (non-small cell lung cancer), HCC (hepatocellular carcinoma), MM (multiple myeloma), NL (the Netherlands), USA (United States of America), UK (United Kingdom), EU (European Union). * Studies with median age above 70 or separate data for this subgroup.Ranking of outcome categories per study.Higher numbers represent higher priority (range 0–5). Where an outcome category is not assessed, no number appears. CA = conjoint analysis, DCE = discrete choice experiment, AHP = Analytic hierarchy process, OPT = outcome prioritization tool. * Studies with median age above 70 or separate data for this subgroup.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01148.txt

@@ -0,0 +1 @@
+Human cytomegalovirus (HCMV) infects 40–70% of adult populations in developed countries and this is thought to be involved in breast cancer progression; however, reports of detection of the viral genome in breast tumors ranges from 0–100%. We optimized a method that is both sensitive and specific to detect HCMV DNA in tissues from Canadian breast cancer patients. Only ~42% of HCMV-seropositive patients expressed viral DNA in their breast tumors. Viral transcription was not detected in any HCMV-infected breast tumors, indicating a latent infection; however, HCMV seropositivity and the presence of latent infections in breast tumors were independently, and in combination, associated with increased metastasis. HCMV DNA-positive tumors were also associated with lower relapse-free survival. Therefore, HCMV infection status should be accounted for during the monitoring and treatment of breast cancer patients. Prevention or reducing the effects of HCMV infection could decrease morbidity and mortality from metastatic disease.Human cytomegalovirus (HCMV) infects 40–70% of adults in developed countries. Detection of HCMV DNA and/or proteins in breast tumors varies considerably, ranging from 0–100%. In this study, nested PCR to detect HCMV glycoprotein B (gB) DNA in breast tumors was shown to be sensitive and specific in contrast to the detection of DNA for immediate early genes. HCMV gB DNA was detected in 18.4% of 136 breast tumors while 62.8% of 94 breast cancer patients were seropositive for HCMV. mRNA for the HCMV immediate early gene was not detected in any sample, suggesting viral latency in breast tumors. HCMV seropositivity was positively correlated with age, body mass index and menopause. Patients who were HCMV seropositive or had HCMV DNA in their tumors were 5.61 (CI 1.77–15.67, p = 0.003) or 5.27 (CI 1.09–28.75, p = 0.039) times more likely to develop Stage IV metastatic tumors, respectively. Patients with HCMV DNA in tumors experienced reduced relapse-free survival (p = 0.042). Being both seropositive with HCMV DNA-positive tumors was associated with vascular involvement and metastasis. We conclude that determining the seropositivity for HCMV and detection of HCMV gB DNA in the breast tumors could identify breast cancer patients more likely to develop metastatic cancer and warrant special treatment.Breast cancer is the most common malignancy among women worldwide and it is their leading cause of death from cancer [1]. Examples of the risk factors for breast cancer development include aging, family history, genetic mutations, reproductive factors such as pregnancy and menopause, endogenous and exogenous effects of estrogen, unhealthy lifestyle such as drinking and smoking, high body-mass index and dense breast tissue [2,3,4,5,6]. Another potential risk factor for the development and progression of breast cancer is viral infection.Infectious agents contribute to >15% of all cancers and ~64% of these agents are viruses [7]. Viruses with causative effects in cancer include the Epstein–Barr virus leading to Burkitt’s lymphomas [8,9] and human papillomavirus leading to cervical cancer [10,11]. Some recent studies also show that human cytomegalovirus (HCMV) has oncogenic potential where it directly leads to oncogenesis [12,13,14]. Indeed, the virus is found in cancerous tissues, including colorectal [15], prostate [16], ovarian [17] and glioblastoma cancers [18]. Increasing evidence also shows that HCMV infection plays a role in breast cancer [19,20,21]. This may be oncogenic [12,13,14] or oncomodulatory where the infection contributes to cancer progression through the alteration of intracellular signaling pathways [22,23].CMV is a β-herpesvirus with double-stranded DNA and infection is highly host specific [24]. HCMV produces lifelong infections in 40–70% of adult populations in developed countries [25,26] and this increases to 85–90% when people reach 75–80 years of age [27]. A higher HCMV seroprevalence is observed in South America, Africa and Asia, where >80% of the non-elderly adult population is infected [28]. Transmission of HCMV is commonly through direct contact with body fluids such as blood, saliva, breast milk and organ transplantation, into which infectious viral particles can be shed during an active infection [29,30,31]. HCMV infection normally causes no noticeable symptoms in immunocompetent individuals; however, it can be associated with fever, malaise, abnormal liver function and infectious mononucleosis [32,33,34]. After active replication of the virus, the infection will enter a state of latency where there is restricted expression of viral genes and limited viral production [35]. Viral reactivation and production of an infectious virus can occur during inflammation, stress, in the aging population, and in immunocompromised hosts [36,37,38,39]. The infected individual is seropositive for HCMV immunoglobulin G (IgG) and the latency stage ensures that this virus is never fully eliminated [40].The presence of HCMV proteins, DNA and/or mRNA in patient breast tumors have been examined by several investigators, but the results are controversial. HCMV immediate early (IE), early and/or late proteins are detected in 90–100% of human breast tumors in some studies [19,41,42,43,44], but at lower incidences in others [45,46,47]. Studies that detect HCMV proteins in a high proportion of breast tumors also show >90% detection of HCMV DNA in these same tumors [19,41,42,43], with two studies also detecting mRNA for HCMV IE [42,43]. By contrast, some studies show a 40–80% positivity of HCMV DNA detection [19,48,49,50], while others report very few or no HCMV DNA-positive breast tumors [45,47,51,52,53,54,55,56,57]. These differences could relate to the variability in serostatus for HCMV in the populations studied, which ranges from 60–100% [45,52,57,58]; however, in a population that is >90% HCMV seropositive, the detection of HCMV DNA in breast cancerous tissues is found to be as low as <20% [45,57]. Alternatively, these discrepancies in the detection of HCMV in breast tumors could also be due to variability in the cellular composition of the breast tumors. We recently showed that the incidence of HCMV infection of breast cancer cells is considerably less than that of fibroblasts with higher infection dependent on the expression of platelet-derived growth factor receptor alpha (PDGFRα) [59]. Fibroblasts are a component of the microenvironment and they are likely to be infected more than the breast cancer cells.Several methods were also used for the detection of HCMV DNA in breast tumors, including in situ hybridization and real-time, standard or nested PCR. The use of these different detection methods, the detection of different viral genes and the source of the starting material such as fresh-frozen or paraffin-embedded tissues could offer another explanation for these disparities [60]. Therefore, it is critical to identify the optimum assay for a chosen viral gene that is both sensitive and specific for detecting HCMV DNA.The association between HCMV infection and outcomes from breast cancer is also important. HCMV DNA and/or proteins in breast tumors have been associated with higher tumor grade [46,49], invasive breast cancer [49], and negative expression of the estrogen receptor-1 (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) [42,46]. Positive detection of HCMV DNA in breast tumors is also associated with poor overall and low relapse-free survival [61]. HCMV DNA and/or proteins are present in nearly all the metastatic lymph nodes [41,43] and brain tissues [62] of breast cancer patients. Our previous work with mouse models of breast cancer showed that the latent infection of mice with mouse cytomegalovirus (mCMV) increases lung metastases, where mCMV DNA, but not mRNA, was detected in the lung metastatic nodules [63]. These combined results indicate that latent CMV infection is associated with worse patient outcomes; however, the relationship between HCMV seropositivity and the presence of HCMV DNA in breast tumors is unclear because the serostatus of patients in previous studies was either unknown or almost all patients were infected. Understanding this relationship and how each of these parameters correlate with patient outcomes is especially important because of the prevalence of the population infected with HCMV, which reaches 40–70% even in developed countries [28]. Testing for seropositivity only requires a blood test and could be useful for directing increased monitoring.In the present study, we assessed two HCMV DNA detection methods, LightCycler (LC)-PCR and nested PCR. Nested PCR targeting the HCMV glycoprotein B (gB) gene provided the required sensitivity and specificity for detecting HCMV DNA in patient breast tumors. Using this method, no significant difference in the detection of HCMV DNA was found in 136 patient breast tumors obtained from a local Canadian population, compared to 10 breast tissue samples from women not diagnosed with breast cancer. The number of breast tumors, which were positive for HCMV DNA, was much lower than the HCMV IgG seropositivity rate in these patients. Notwithstanding, HCMV seropositivity and the presence of gB DNA in the breast tumors were each positively associated with metastasis. Patients with HCMV gB DNA detected in the breast tumors were also at higher risk of having a lower relapse-free survival time. In addition, mRNA of HCMV IE1 was not detected in the HCMV gB DNA-positive breast tumors, indicating a presumably latent infection. Our results, therefore, demonstrate that breast cancer patients who are HCMV seropositive need more intensive monitoring for disease progression, particularly those patients where the excised tumors are positive for HCMV DNA.All common chemicals and reagents were purchased from Sigma-Aldrich (Oakville, ON, Canada) or Thermo Fisher Scientific (Waltham, MA, USA) unless otherwise stated. All primers were from Integrated DNA Technologies Inc. (Coralville, IA, USA).Breast tumors from 136 breast cancer patients were obtained after surgical removal and they were immediately frozen in liquid nitrogen, followed by storage at −80 °C by the Alberta Cancer Research Biobank from where they were obtained. Normal breast tissues from ten women who had no history of cancer were obtained from breast reduction surgeries and processed similarly as the breast tumors. Patient samples were obtained with the approval of the University of Alberta Health Research Ethics Board (Pro00018758) with written informed consent.Matching patient information was available for each breast tumor, including age, menopause status, body mass index, tumor grade, number and size of lymph node metastases, tumor stage, expression of estrogen receptor (ER)/progesterone receptor (PR)/human epidermal growth factor receptor 2 (HER2) in the tumor, tumor recurrence, vascular invasion and patient survival. All information was updated in November 2020. Cases with unknown or missing information in each category were excluded from the statistical analysis for that category. Matching serum samples for 84 of these breast cancer patients were available for serological tests.A clinical isolate of HCMV, Kp7 (from Dr. Jutta K. Preiksaitis, Department of Medicine, University of Alberta) [64], was used to generate the HCMV-positive control. Freshly obtained human breast adipose tissue from a breast reduction surgery that initially tested negative for HCMV DNA was actively infected with 1.25 × 104 virus/mL of HCMV in culture for 24 h and then extracted for DNA and RNA. The extracted DNA was serially diluted 10-fold (neat to 10−5). The HCMV-negative control was prepared from human MDA-MB-231 breast cancer cells (American Type Culture Collection, Manassas, VA, USA), where the extracted DNA and mRNA were both negative for HCMV.DNA and RNA were extracted using the All-In-One DNA/RNA/Protein Miniprep Kit (Bio Basic, Markham, ON, Canada) following the manufacturer’s protocol. Briefly, tissues (~20 mg) were homogenized in a 350 μL Buffer Lysis-DRP contained in 2 mL microcentrifuge tubes with 5-mm stainless steel beads, using the Qiagen TissueLyser II system (24-sample plates, 25 Hz, 5 min) (Qiagen, Toronto, ON, Canada). Tissue lysates were centrifuged at 12,000× g for 3 min at 4 °C and the supernatants were transferred into new RNase-Free tubes. Lysates were transferred into the EZ-10 DNA Columns and centrifuged at 9000× g for 1 min at room temperature, with the columns kept for DNA purification at room temperature and the flow-throughs transferred to new RNase-Free tubes for RNA purification at 4 °C.For DNA purification, the EZ-10 DNA columns were washed sequentially by 350 μL Buffer Lysis-DRP, 500 μL CW1 Solution and 500 μL CW2 Solution. Each solution was added to the columns and incubated for 1 min, then centrifuged at 9000× g for 1 min before the next solution was added. The columns were further centrifuged at 9000× g for 2 min to remove residual liquid. The columns were opened for 3 min to evaporate the ethanol, then incubated with 50 μL CE Buffer for 2 min and centrifuged at 9000× g for 2 min to elute the DNA. Samples were stored at −20 °C.For RNA purification, 250 μL ethanol was added to the previously collected flow-throughs, with the mixture transferred into the RZ-10 RNA columns and centrifuged at 9000× g for 1 min. The second flow-throughs were kept for protein purification. The RZ-10 RNA columns were washed sequentially by 500 μL GT Solution and 500 μL NT Solution. Each solution was added to the columns and incubated for 1 min, then centrifuged at 9000× g for 1 min before the next solution was added. The columns were further centrifuged at 9000× g for 2 min to remove the residual solution. The columns were incubated with 50 μL RNase-Free Water for 2 min and centrifuged at 9000× g for 2 min to elute the RNA. Samples were stored at −80 °C.The LC-PCR reaction mixture contained 5 μL of DNA, 4 mM MgCl2, 0.5 μM of each primer, 0.2 μM of each probe, and 2 μL of the reagent from a LC-FastStart DNA Master hybridization probe kit (Roche Diagnostics, Laval, QC, Canada) for a total reaction volume of 20 μL. The primers used for the detection of HCMV gB were as follows: forward: 5′-TACCCCTATCGCGTGTGTTC-3′ and reverse: 5′-ATAGGAGGCGCCACGTATTCT-3′. The hybridization donor probe with a fluorescein 3′-end label and the acceptor probe with a LC-Red 640 5′-end label were used in the LC-PCR reaction (TIB Molbiol LLC, New Jersey, NJ, USA) [65]. The reaction was performed in the LightCycler 480 Instrument II (Roche Diagnostics, Laval, PQ, Canada), with the following thermal cycling conditions: (1) 10 min at 95 °C and (2) 45 cycles of 15 s of denaturing at 95 °C, 10 s of annealing at 55 °C, and 10 s extension at 72 °C. Measurements were collected during the annealing period with a channel setting F2/F1 for real-time detection of the amplification. The specificity of the fluorescence signal was checked by a melting curve analysis, with Tm = 67.5 °C for the probes.A positive standard curve was generated with a series of log dilutions containing 106 to 101 genome copies of HCMV, using purified viral DNA that was quantified by spectrophotometry with absorbance at 260 nm [65]. The LC-PCR program compared the results from the tested samples against the standard curve to determine the number of HCMV genome copies in a sample reaction.The presence of HCMV DNA in human tissues was determined using primers specific to either HCMV IE1 (fourth exon) [58,66] or gB (UL55) gene [66]. Each PCR reaction had a 50 μL reaction volume, containing 2X PCR Taq Master Mix (Applied Biological Materials, Richmond, BC, Canada). The extracted DNA (150 ng) was amplified with the external primer set for the first round of amplification, then 2 μL of the amplified product was used with the internal primer set for the second round of amplification. The PCR reaction was performed with an initial denaturation for 4 min at 94 °C, followed by specific thermal cycling conditions that were individually listed for each primer set in Table S1 in the order of denaturation, annealing and elongation, and then a final elongation for 7 min at 72 °C. Optimization of the PCR conditions was performed for the HCMV IE1 primers by testing a range of annealing temperatures. For the external primers, annealing temperatures of 55, 60, 62, 65 and 67 °C were tested, with all second rounds of reaction performed at 50 °C. The internal primers were tested at annealing temperatures of 50, 53 and 55 °C, after the first round of reaction was performed at 62 °C. The products were visualized after separation on an agarose gel, staining with ethidium bromide and exposure to ultraviolet light. For products > 100 base pair (bp), a 1.5% gel was used, and electrophoresis was performed at 100 V for 30 min. For products < 100 bp, a 3% gel was used and developed for 1 h at 100 V. The band size was determined relative to the 100 bp DNA ladder. All steps were performed with care to avoid cross contamination between samples and new aliquots of reagents were used for each round of PCR reaction.PCR products with the expected size were cut out from the agarose gel for re-extraction of DNA using the MinElute Gel Extraction Kit (Qiagen, Toronto, ON, Canada) following the manufacture’s protocol. PCR products < 100 bp were cloned into vectors for effective DNA sequencing. Cloning was performed using the pUCM-T Cloning Vector Kit (Bio Basic Inc., Markham, ON, Canada) and in competent E. coli cells. The resulting colonies were screened for transformants, with the white colonies representing recombinant clones. Each colony was picked up with a toothpick and placed in 5 mL of LB medium containing ampicillin for overnight incubation at 37 °C. Plasmid DNA was extracted from the liquid culture using the Column-Pure Plasmid Miniprep Kit (Applied Biological Materials, Richmond, BC, Canada). Samples were analyzed using the Sanger DNA Sequencing service at the Molecular Biology Facility (University of Alberta, Edmonton, AB, Canada). The sequence of the target clones was determined by M13 universal forward primer: 5′-GTAAAACGACGGCCAGT-3′. The resulting sequences were aligned to the target gene using the CLC Sequence Viewer software (CLC bio, Aarhus, Denmark) to verify the product specificity. The sequences were further identified using the Basic Local Alignment Search Tool (BLAST) database.mRNA was reverse transcribed to complementary DNA (cDNA) using the 5X All-In-One Reverse Transcription MasterMix (Applied Biological Materials, Richmond, BC, Canada). The cDNA was analyzed for the relative amount of target genes by qPCR using EvaGreen qPCR master mix (Applied Biological Materials, Richmond, BC, Canada). The relative abundance of HCMV IE1 [67] and PDGFRα expression at the mRNA level was determined by normalizing against a housekeeping gene, glyceraldehyde 3-phosphate dehydrogenase (GAPDH). HCMV IE: sense 5′-TGAGGATAAGCGGGAGATGT-3′ and antisense 5′-ACTGAGGCAAGTTCTGCAGT-3′. PDGFRα: sense 5′- TAGTGCTTGGTCGGGTCTTG -3′ and antisense 5′- TTCATGACAGGTTGGGACCG -3′. GAPDH: sense 5′-TCCTGCACCACCAACTGCTT-3′ and antisense 5′-TCTTACTCCTTGGAGGCCAT-3′.The presence of HCMV IgG antibodies in serum samples was determined qualitatively by using the CMV IgG ELISA kit (Genway Biotech Inc., San Diego, CA, USA) according to the manufacturer’s instructions. Samples were added to microtiter strip wells precoated with CMV antigens, allowing the binding of the CMV IgG antibodies to the well. Wells were washed and horseradish peroxidase (HRP) labelled anti-human IgG conjugate was added. Tetramethylbenzidine, the substrate for HRP was then added, giving a blue product when CMV IgG antibodies were present in the sample. Sulfuric acid was added to stop the reaction, resulting in a yellow endpoint color, with the absorbance determined at 450 nm using Easy Reader EAR 340 AT (SLT-Lab Instruments, Salzburg, Austria). Appropriate controls were included as suggested by the manufacturer to determine the positivity of tested samples.Descriptive statistics were used to present the study variables. The mean and S.D. or median and interquartile range (IQR) were reported for continuous variables, based on a normal or skewed distribution of the results, respectively. Frequency and proportions were reported for categorical variables. Chi-square tests were used to correlate two categorical variables. Fisher’s exact tests were reported when the cell frequency was less than 5. Binary logistic regression was used to identify the factors associated with the gB DNA (negative versus positive) outcome variable. A univariate binary logistic regression model was used to find the variables associated with the outcome variable. Factors significant at p < 0.10 were entered into the multivariate model. The final multivariate model was chosen based on the statistical significance and clinical relevance. A p-value < 0.05 was used for statistical significance. SPSS (IBM Corp. Released 2017. IBM SPSS Statistics for Windows, Version 25.0. Armonk, NY, USA: IBM Corp.) version 25 was used for all statistical analysis.We compared the limit of detection for HCMV gB DNA as measured by LC-PCR and nested PCR in parallel. The HCMV DNA-positive control including the original DNA extract and dilutions at 1:10, 1:102, 1:103, 1:104 and 1:105 were tested. The HCMV DNA-negative control was included in each test. The LC-PCR analysis resulted in an estimate of 200 copies of the HCMV viral genome in 10 μL of the original stock of HCMV DNA-positive control based on calculations using a previously established standard curve that was generated with purified HCMV DNA of known genome copies (Table 1) [62]; however, 104 copies of the viral genome were estimated to be present in 10 µL of the 1:10 diluted HCMV DNA-positive control instead of an expected 20 copies (Table 1). LC-PCR was able to detect HCMV gB DNA up to 1:102 dilution of the HCMV positive control, with an estimation of two copies of the viral genome in the reaction, while 1:103, 1:104 and 1:105 dilutions were all below the level of detection (Table 1). By contrast, the nested PCR detected a band at 96 bp on the gel, representing the HCMV gB gene, even when the HCMV-positive control was diluted 1:105 (Figure 1). This demonstrates that the nested PCR had an ~1000-fold greater sensitivity in comparison to the LC-PCR; however, different dilutions of the HCMV-positive control resulted in no obvious difference in the intensity of bands, representing saturation of the reactions, where observation of a band would only indicate the presence of the viral gene, but not its level (Figure 1). Both methods showed no detection of HCMV gB DNA in the HCMV-negative control sample (Table 1, Figure 1 and Figure 2). Furthermore, no non-specific bands were observed for the nested PCR product of HCMV gB in the negative HCMV control, indicating that the reaction was specific (Figure 1).The presence of the HCMV viral genome in human breast tumors and normal breast tissues was examined by nested PCR that targeted either the gB or IE1 genes, using a subset of the total samples to confirm the methods. The HCMV-positive (1:103 dilution) and negative controls were included in every round of PCR amplification. A reaction mixture with no template was also included to monitor contamination during the experimental procedure. No bands were observed on the gel for any of these negative controls.The PCR products generated by targeting the HCMV gB gene in human breast tumor samples showed no band or a single band of similar intensities at 96 bp similar to the positive control (Figure 2A). The identity of this band as CMV gB DNA was verified by Sanger sequencing of ten clones created from bands extracted from the gels. All ten sequences showed alignment to the UL55 genomic sequence that encodes HCMV gB. Four of these clones are illustrated as examples in the sequence alignment map (Figure 2B). This demonstrates that the 96 bp PCR products were not false positives.On the other hand, the detection of HCMV IE1 DNA in human breast tumor samples showed ambiguous results. As expected, the HCMV-positive control showed a single band at 293 bp on the gel (Figure 3A). Multiple bands were observed in the HCMV negative control, although the products were <200 bp (Figure 3A). The human breast tumor samples showed highly variable results with multiple bands. To increase the primer specificity a gradient of annealing temperatures was tested in the PCR reaction, up to a 5 °C increase in either the external or internal primer reactions, but this still resulted in multiple bands for the human breast tumor samples tested (Figure S1). Some samples appeared to have a product of 293 bp, but this was not the most intense band (Figure 3A). Seven patient samples were chosen randomly, with the observed ~293 bp bands re-extracted for the DNA products. These samples were sequenced but none aligned with the IE1 genomic sequence (Figure 3B). The resulting sequences were identified using the BLAST database and all were found to be from the human genomic sequence. Therefore, the detection of HCMV IE1 using nested PCR produced non-specific products and false-positive results.Nested PCR for the detection of HCMV gB DNA was specific and it was, therefore, used to analyze all tissues. Tissue samples from breast reduction surgeries from ten women without breast cancer and 136 breast tumors from patients were analyzed resulting in 10% (1/10) and 18.4% (25/136) positivity for HCMV gB DNA, respectively. These results were not significantly different (p = 0.691, Table 2). Of the 84 matched serum samples that were available from the breast cancer patients, 41.7% (35/84) were negative and 58.3% (49/84) were positive for HCMV IgG antibody (Table 2). All patients who were seronegative for HCMV IgG were also negative for HCMV gB DNA in the breast tumors, thus confirming the lack of infection. Conversely, all women who had HCMV gB DNA positive tumors were also positive for serum HCMV IgG. There were ten patients with HCMV gB DNA detected in the breast tumors but for whom matching serum samples were not available. Therefore, for further analyses, we assumed that these ten patients with breast tumors would have also been seropositive for HCMV. As a result, we calculated that 62.8% (59/94) of our patient population were HCMV seropositive and 37.2% (35/94) were HCMV seronegative (Table 2). Including the ten patients with assumed HCMV seropositivity, 42.4% (25/59) of the HCMV IgG positive patients were also positive for HCMV gB DNA in breast tumors, while no HCMV gB DNA was detected in 57.6% (34/59) of HCMV seropositive women (Table 2).We next examined whether the presence of HCMV gB DNA in breast tumors was associated with any patient characteristics. The average age of all patients was 56, ranging from 23 to 80 years (Table 3). Most patient characteristics did not differ between groups; however, out of all patients that had tumors positive for HCMV gB DNA, 64.0% (16/25) had a tumor recurrence event, which was significantly higher compared to the 43.2% (48/111) of patients in the HCMV gB DNA-negative group in univariate analysis (p = 0.06, Table 3). In addition, 32.0% (8/25) of patients with HCMV gB DNA-positive tumors had Stage IV or metastatic breast cancer, which was significantly higher than the 10.8% (12/111) for the HCMV gB DNA-negative group (p = 0.007, Table 3). These results showed that HCMV gB DNA positivity in breast tumors was positively associated with tumor recurrence events and metastasis.We also examined whether being HCMV seropositive, regardless of gB DNA status in breast tumors, was associated with any patient characteristics by univariate analysis. Patients who were positive for HCMV IgG were significantly older at the time of breast cancer diagnosis (57 ± 12 years) compared to those that were HCMV IgG negative (51 ± 13 years) (p = 0.017, Table 4). A higher percentage of HCMV IgG negative patients (50.0%, 17/34) were pre-menopausal compared to those that were IgG positive (21.1%, 12/57). Correspondingly, a higher percentage of HCMV IgG positive patients were post-menopausal (68.4%, 39/57) and 10.5% (6/57) were pre-menopausal compared to those that were IgG negative (47.1%, 16/34 or 2.9%, 1/34, respectively). Overall, HCMV IgG seropositivity was significantly associated with post- and peri-menopause (p = 0.012, Table 4). In addition, patients who were positive for HCMV IgG had a significantly higher body mass index (29.9 ± 7.36 kg/m2) compared to those that were HCMV IgG negative (27.5 ± 6.02 kg/m2) (p = 0.093, Table 4); however, age, menopause status and body mass index were not significantly associated with the presence of HCMV gB DNA in the breast tumors (Table 3).Stage IV or metastatic breast cancer was significantly higher in patients who were HCMV IgG seropositive with HCMV gB DNA detected in breast tumors (32.0% (8/25)) compared to patients that were seropositive but HCMV gB DNA-negative (11.8% (4/34)) when assessed by univariate analysis (p < 0.0995, Table 5). Similarly, IgG seropositive patients with detection of gB DNA in the breast tumor had significantly increased vascular involvement (70.8% (17/24)) compared to IgG-positive patients with gB DNA-negative tumors (45.5% (15/33)) (p = 0.057, Table 5); however, vascular involvement was not associated with either HCMV seropositivity or gB DNA-positive tumors on their own (Table 3 and Table 4).Other patient outcomes and characteristics including overall survival status, overall time to recurrence event, tumor grade, number and size of metastatic lymph nodes, expression of ER/PR/HER2, tumor subtypes or tumor PDGFRα mRNA expression were not significantly different when assessing patients based on the presence or absence of HCMV gB DNA in breast tumors or based on HCMV serostatus (Table 3, Table 4 and Table 5).Based on these results from the univariate analyses, we next performed multivariate analysis. The odds of being HCMV gB DNA positive in breast tumors or being HCMV IgG positive regardless of gB DNA status in patients with Stage IV breast cancer was 5.27 or 5.61 times higher, respectively, than patients with Stage I-III (95% CI 1.77–15.67, p = 0.003, 95% CI 1.09–28.75, p = 0.039; Table 6). HCMV gB DNA positivity was not significantly associated with menopause status even after adjustment (Table 6); however, when compared to pre-menopausal patients, the odds of being HCMV IgG positive regardless of gB DNA status for post-menopause patients was 3.83 times higher (95% CI 1.43–10.29, p = 0.008) and for peri-menopause patients was 11.42 times higher (95% CI 1.18–110.31, p = 0.035, Table 6). Within the HCMV seropositive group, the odds ratio of having gB DNA-positive breast tumors and being peri- or post-menopausal were not significantly different compared to seropositive patients with gB DNA-negative tumors (Table 6; however, there was a trend towards an increased odds ratio of developing Stage IV breast cancer in seropositive patients with gB DNA-positive tumors compared to those who were gB DNA negative (3.48, 95% CI 0.88–13.78, p = 0.076).We examined the association between HCMV gB DNA-positive tumors and survival time. Median overall survival time in patients who were HCMV gB DNA positive (7.06 ± 2.17 years, 95% CI 2.81–11.3) compared to those that were negative (8.72 ± 1.36 years, 95% CI 6.05–11.4) was not significantly different (p = 0.614, Figure 4A). Patients that had HCMV gB DNA-negative or -positive tumors had 64.4% versus 59.8% overall survival rates at the five-year interval and 45.6% versus 45.0% at the ten-year interval, respectively (Figure 4A); however, the median relapse-free survival time for patients with HCMV gB DNA-positive tumors (3.45 ± 1.61 years, 95% CI 0.29–6.61) was significantly lower than for those with HCMV gB DNA-negative tumors (8.51 years, 95% CI not reached) (p = 0.039, Figure 4B). Patients with HCMV gB DNA-negative or -positive tumors had 58.3% versus 33.6% relapse-free survival rates at the five-year interval and 49.1% versus 28.8% at the ten-year interval, respectively (Figure 4B). The risk of reduced relapse-free survival was 1.8 times higher in patients who were positive for HCMV gB DNA in their tumors compared to those whose tumors were negative (95% CI 1.02−3.17, p = 0.042, Figure 4B).Neither overall nor relapse-free survival were different in patients based solely on HCMV serostatus (Figure S2). The five-year and ten-year overall survival rates for HCMV IgG-negative patients were 70.9% and 53.6%, respectively, and for IgG-positive patients they were 64.1% and 51.6%, respectively. The five-year and ten-year relapse-free survival rates for the HCMV IgG-negative patients were 63.2% and 58.3%, respectively, and for the HCMV IgG-positive patients were 46.3% and 39.1%, respectively. The risk of reduced relapse-free survival was 1.55 times higher in the HCMV IgG-positive patients, but this did not reach statistical significance (95% CI 0.81−2.96, p = 0.185, Figure S2).The expression of HCMV IE1 mRNA was examined in the same set of human breast tumors (n = 136) and normal breast tissues (n = 10). The HCMV-positive and -negative controls were included in each round of qPCR analysis. The HCMV-positive control resulted in an amplification cycle number of ~22 and a single peak in the melting curve at 86.5 °C (Figure S3). All of the patient samples tested showed undetermined or >35 amplification cycle numbers (Figure S3A). The samples that showed a value for amplification did not show a single peak that matched with the positive control on the melt curve (Figure S3B). These results indicated that the mRNA expression of HCMV IE1 was not detected in any of the patient samples analyzed, even if HCMV gB DNA was present. This indicates that the HCMV infection was latent in these tumors.Increasing evidence shows that HCMV infection is associated with breast cancer and metastasis [20,21]; however, the reported rate of positive detection of HCMV DNA/mRNA/proteins in breast tumors is highly variable in studies performed around the world, which ranges from 0 to 100% [19,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,61,62]. While this discrepancy could be partly caused by differences in active infection levels in different countries, it is also likely that the use of different detection methods could be a contributing factor. We compared real-time and gel-based PCR techniques for the detection of HCMV DNA and determined that a nested PCR, targeting the HCMV gB gene, was both sensitive and specific for analyzing human breast tumors. In contrast, using this same technique to detect the HCMV IE1 gene led to non-specific results.We therefore used nested PCR to measure HCMV gB DNA in breast tumors from a local Canadian patient population in addition to assessing the available matched serum samples for HCMV seropositivity. Patients with metastatic tumors were 5.61 or 5.27 times more likely to be HCMV seropositive or to have HCMV gB DNA-positive tumors, respectively, while the latter also had decreased relapse-free survival.Detection of HCMV DNA in tissues normally relies on PCR; however, the reported results are highly variable. One explanation could be the variability and specificity of the methods used. Real-time PCR using the LightCycler, quantifies the target gene, while standard and nested PCR only detect the presence of the end products on a gel. We compared LC-PCR and nested PCR directly for the detection of HCMV gB DNA in our patient samples. Strict precautionary measures were taken when both PCR techniques were performed to avoid possible cross-contamination. Although the LC-PCR was very specific with the use of hybridization probes and allowed for real-time quantitative analysis of the viral load, nested PCR showed a 1000-fold higher sensitivity; however, the results from the nested PCR only indicated the presence of the viral DNA, not the amount, which limits information about the infection status. Several studies showing very little or no detection of HCMV DNA in breast tumors used detection methods similar to LC-PCR, which involved real-time analysis that was not the most sensitive detection method in our hands [45,47,51,52,53,54,55]; meanwhile, two other studies that used real-time PCR detected HCMV IE DNA in all 12 and 146 patient samples tested ([41,43], respectively). These same samples were all positive by IHC for HCMV IE or L proteins, which indicates an active infection with a high level of virus that likely exceeded the detection limit of real-time PCR [41,43]. In fact, one of these studies also showed the detection of HCMV mRNA in the tested tumors [43].DNA detection must be both sensitive and specific to avoid false-positive results. In comparison to standard PCR, nested PCR uses two different sets of primers with one nested in the other and also uses two rounds of PCR reactions to enhance the amplification while ensuring its specificity. To confirm specificity, we performed Sanger DNA sequencing on the nested PCR products and demonstrated that only the expected product was obtained for HCMV gB. By contrast, nested PCR products obtained from targeting the HCMV IE1 gene were not specific. This was surprising because the same HCMV IE1 primers and PCR conditions were described previously to be specific when the PCR products were sequenced [58]. El-Shinawi et al. detected HCMV IE1 DNA in 53.1% (26 out of 49) of tumors from non-inflammatory breast cancer patients with no detection in normal breast tissues from women free of cancer (the sample number was not specified) [58]. In a later study with more patients, the same authors reported HCMV IE1 DNA in 74% (67 out of 91) of non-inflammatory breast cancer patients and 89% (39 out of 44) of inflammatory breast cancer patients [50]. Although we followed the exact PCR detection method described in these studies, our PCR products were non-specific and resulted from the amplification of human genome sequences. We also tried to optimize the PCR reaction by testing a gradient of annealing temperatures based on what was reported, since the efficiency of PCR machines may be different; however, this did not decrease the non-specific products. A possible explanation for this discrepancy is the mismatch between the primers used in each study and the local circulating strains of HCMV. Genome variability of HCMV has been reported in congenitally infected infants [68], with the IE1 gene potentially evolving, since it is a common target for CD8+ T cell responses [69]. Another possible explanation is that the HCMV DNA amount in the samples differed between their studies and ours, where non-specific reactions could overpower the target gene amplification if the level of viral genome present was very low. This is likely since the gels illustrated by El-Shinawi et al. also showed some non-specific bands, although these were very faint compared to the target gene product [58]. This suggestion is also supported in our study, since the HCMV positive control generated after in vitro infection of the human breast tissue, contained a high level of viral DNA that was able to produce a single strong band on the gel at the expected position. Nevertheless, the use of nested PCR to detect HCMV IE1 DNA was not a reliable method for analyzing patient samples in our hands. Instead, we used nested PCR targeting HCMV gB DNA, which was both sensitive and specific.Previous studies showed a significantly higher percentage of breast tumors positive for HCMV DNA and/or proteins compared to normal breast tissues [19,43,44,45,57]. This could provide evidence that HCMV fulfills one of the criteria for being an oncovirus [70]; however, two other groups reported no significant differences in the detection of HCMV DNA in breast tumors versus normal breast tissues [48,52], which are compatible with our results. In our study, a high prevalence of HCMV DNA in breast tumors compared to normal breast tissues was not observed, although the low sample size for normal breast tissues could be statistically limiting. Compared to the seropositivity rate of 63% determined for this breast cancer patient population, HCMV gB DNA was only detected in 18% of the breast tumors. This result is compatible with other published studies [45,57]. Additionally, in our studies with mice that were latently infected with mCMV, we could only detect mCMV DNA in one or two but not all tissues examined including salivary gland, spleen, kidney, lung and breast tumors [63,71]. It is possible in this work that viral DNA was present in more tissues, but that it was below the level of detection.We found no evidence for active or reactivated HCMV infections in the HCMV DNA-positive breast tumors or in breast tissue from women without cancer as shown by the lack of mRNA expression for HCMV IE, indicating that infection in these tissues was latent. This is surprising since the reactivation of CMV infection can be induced by inflammation, and we showed previously that there is a proinflammatory condition at the breast tumor site and in the adjacent adipose tissues [72]. Additionally, reactivation of CMV often occurs in differentiated macrophages and dendritic cells, which are present in breast tumors [73,74,75]. The seropositivity rate of 63% HCMV in our Canadian breast cancer patients is within the expected 40–70% infection rate for the adult population in North America [26]; however, this seropositivity is lower than in other countries where almost all of the patients are infected as determined through HCMV IgG testing [45] or the presence of HCMV DNA in the breast tumors [19,41,42,43,44]. In fact, HCMV mRNA was detectable in breast tumors in studies where 100% of the tumors examined were positive for HCMV DNA and/or proteins [42,43]. It is possible that there is a higher reactivation rate leading to the detection of mRNA in countries with very high HCMV infection rates. It is also possible that there was HCMV mRNA in the breast tumors from our patient population, but that the expression was very low and below the level of detection.An important finding of this present study is that women who were seropositive for HCMV IgG regardless of the presence of gB DNA in their breast tumors were more likely to develop Stage IV metastatic breast cancer (OR 5.61, CI 1.09–28.75, p = 0.039). This was also true for breast cancer patients with HCMV gB DNA-positive tumors compared to all other patients with gB DNA-negative tumors regardless of HCMV status even without evidence of an active HCMV infection. More significantly, women who were HCMV seropositive and whose breast tumors were gB DNA-positive were more likely to develop Stage IV metastatic breast cancer with vascular involvement compared to those women who were HCMV seropositive but gB DNA-negative. These results are compatible with work where HCMV DNA and/or proteins are abundantly expressed in metastatic sentinel lymph nodes [41,43] and brain tissues [62] from breast cancer patients, indicating an association between infection and metastasis. In addition, our preclinical study shows that mice with latent mCMV infection developed more and larger metastatic lung nodules [63].HCMV seropositivity increased with age in our study as expected [27] and was also associated with higher body mass index, which is also a risk factor for breast cancer [2]. A significantly higher incidence of infection was found in post- and peri-menopausal women compared to pre-menopausal women. This should be considered for follow-up and treatment options that are dependent on menopause status. Our study did not measure differences in the levels of HCMV IgG. One study in which almost all patients were HCMV seropositive, shows higher HCMV IgG levels in breast cancer patients compared to the control group [57], while another study shows no differences [76]. As well, HCMV-seropositive women with breast cancer who are less than 40 years of age have higher mean IgG levels than those without breast cancer. Higher IgG levels indicate a primary or reactivated infection that could therefore be considered a risk factor for breast cancer [77].The presence of HCMV gB DNA in breast tumors in the present study was associated with worse outcomes in breast cancer patients. A study with low detection of HCMV DNA in breast tumors at less than 10% shows no viral association with clinical factors [56]. Another study with ~76% HCMV gB DNA detection in breast tumors shows that HCMV infection is associated with poor overall and relapse-free survival [61]. In our study of Canadian women, having HCMV gB DNA-positive breast tumors was associated with a reduced relapse-free survival, overall tumor recurrence and metastasis, but was not associated with reduced overall survival.The presence of HCMV antigens in breast tumors is correlated with the lack of expression of ER, PR or HER2 in some studies [42,46], while others show a relationship with HER2 overexpression in HCMV-positive breast tumors [44]. In our study, HCMV seropositivity or the presence of gB DNA in the breast tumors was not associated with tumor subtypes; however, most tumors in this cohort were hormone-receptor positive. In experiments with cultured cells, we showed that the level of HCMV infectivity of different breast cancer cell lines was much lower than in fibroblasts. This did not depend on whether the breast cancer cells were triple negative or hormone receptor positive [59]. Instead, infection levels depended on the expression of PDGFRα. This receptor facilitates HCMV uptake in epithelial cells [78]. Despite this, there was no significant association between HCMV seropositivity or the presence of gB DNA in the breast tumors and the mRNA expression of PDGFRα in our patient samples; however, the proportion and characteristics of different cell types that constitute the breast tumors of patients could have a major impact on whether the tumors are positive or not for HCMV.This study specifically evaluated two commonly used DNA detection methods for analyzing the presence of the HCMV genome in human breast tumors. Nested PCR was much more sensitive than real-time PCR and it detected DNA for HCMV gB with high specificity. By contrast, detection of HCMV IE1 DNA was not specific because of interference from the human genome. Using nested PCR, 18% of tumors from our Canadian breast cancer patients were positive for HCMV gB DNA whereas 63% of the breast cancer patient patients were seropositive for HCMV. HCMV seropositive patients were more likely to be older and to include post- and peri-menopausal women or those with a high body mass index. We directly related HCMV seropositivity to Stage IV metastatic breast cancer for the first time. HCMV infection in breast tumors was presumed to be latent because although viral DNA was detected, viral mRNA was not. These results are compatible with our mouse studies where latent mCMV infection increased the size and number of lung metastases [63]. Our present study shows that determining the seropositivity for HCMV and subsequent detection of HCMV gB DNA in breast tumors could identify breast cancer patients who are more likely to develop metastatic cancer and warrant special treatment to increase their relapse-free survival.The following are available online at https://www.mdpi.com/article/10.3390/cancers14051148/s1, Figure S1: HCMV immediate early 1 (IE1) DNA detection using nested PCR with different annealing temperatures; Figure S2: Kaplan–Meier survival curve based on HCMV IgG seropositivity; Figure S3: mRNA expression of HCMV immediate early 1 (IE1) gene in human breast tumor samples; Table S1: Primer sequences for nested PCR.Conceptualization, Z.Y., D.N.B. and D.G.H.; methodology, Z.Y., X.T., M.E.H., X.P., S.G., T.P.W.M., D.N.B. and D.G.H.; formal analysis, Z.Y., S.G., D.N.B. and D.G.H.; investigation, Z.Y. and X.T.; resources, T.P.W.M., D.N.B. and D.G.H.; writing—original draft preparation, Z.Y.; writing—review and editing, Z.Y., X.T., M.E.H., X.P., S.G., T.P.W.M., D.N.B. and D.G.H.; visualization, Z.Y., D.N.B. and D.G.H.; supervision, D.N.B. and D.G.H.; project administration, D.N.B. and D.G.H.; funding acquisition, D.N.B. and D.G.H. All authors have read and agreed to the published version of the manuscript.This research was supported by Innovative Grants from the Canadian Cancer Society Research Institute/Canadian Breast Cancer Foundation (300034) and the Women and Children’s Health Research Institute (WCHRI) at the University of Alberta. D.G.H. also received funding from the Canadian Institutes of Health Research (CIHR; MOP123488) and the Li Ka Shing Institute of Virology at the University of Alberta. D.N.B. also received funding from the CIHR (PJT-169140) including publication costs. The authors thank Dr. Kathryn Graham for help in providing patient samples and deidentified records.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Health Research Ethics Board of Alberta Cancer Committee (HREBA.CC-181313; Renewal 3 on 16 May 2021).Informed consent was obtained from all subjects involved in the study.All relevant data for the patients in this study is provided in the Tables, Figures and Supplementary Results. All patient samples were deidentified to preserve patient confidentiality.The authors declare no conflict of interest.The detection limit of LightCycler PCR (LC-PCR) and nested PCR for HCMV glycoprotein B (gB) DNA. HCMV positive control was prepared in serial dilutions ranging from 1 to 1:105. The samples were examined for the presence of HCMV gB DNA using LC-PCR (Table 1) and nested PCR in this Figure. The calculated concentration in Table 1 was determined by the LC-PCR program based on the previously established HCMV standards. Samples illustrated in lanes 1–6 are as follows: 100 base pair (bp) DNA ladder, HCMV 1:103, 1:104, 1:105, negative control, positive control (HCMV 1:10). Undetermined cycle numbers are at least >40 and concentration cannot be calculated. 3.2. Specific Detection of HCMV gB but not HCMV IE1 Using Analysis by Nested PCR of Human Specimens.Specificity of HCMV glycoprotein (gB) DNA detection using nested PCR. PCR products for amplified HCMV gB gene from human breast tumor samples were visualized on 3% agarose gel stained with ethidium bromide, resulting in a band at 96 bp in the positive control and some human breast tumor samples (A). The PCR product at the 96 base-pair (bp) position was re-extracted from the gel and cloned for Sanger DNA sequencing, with the results aligned by comparing to the UL55 genomic sequence that encodes HCMV gB (B). Ten samples were analyzed and all aligned to UL55. Four of the samples were randomly chosen and illustrated in the sequence alignment map (B).Specificity of HCMV immediate early 1 (IE1) DNA detection using nested PCR. PCR products for amplified HCMV IE1 gene from human breast tumor samples were visualized on 1.5% agarose gel stained with ethidium bromide, resulting with a band at 293 base pair (bp) in the positive control (A). The PCR product at the 293 bp position was re-extracted from the gel for Sanger DNA sequencing, with the results aligned to the HCMV IE1 genomic sequence (B). Seven samples were analyzed and illustrated in the sequence alignment map (B).Kaplan–Meier survival curves based on HCMV gB DNA detection in breast tumors. (A) Overall survival and (B) Relapse-free survival. N = 136. p-values were calculated by the Chi-square analysis. p-value was considered significant if < 0.05. (A) p = 0.614 and (B) p = 0.039.HCMV positive control was prepared in serial dilutions ranging from 1 to 1:105. The samples were examined for the presence of HCMV gB DNA using LC-PCR.NA = not available.Detection of HCMV DNA in tissues and IgG in serum of patients.* There were ten patients with HCMV gB DNA-positive tumors who had no matching serum. These patients were assumed to be seropositive for HCMV. p = 0.691 comparing the number of HCMV gB-positive breast tissues to breast tumors was calculated by the Fisher’s exact test. * p value < 0.1 considered significant. NA = not applicable, gB = glycoprotein B.HCMV gB DNA negative and positive breast tumors: univariate analysis of patient characteristics.p-values were calculated by Chi-square analysis or the Fisher’s exact test if n < 5 in a category. p-value was considered significant if <0.1 and these values are bolded. ER: estrogen receptor; gB: glycoprotein B; HER2: human epidermal growth factor receptor 2; PDGFRα: platelet-derived growth factor receptor alpha; PR: progesterone receptor.HCMV IgG seronegative and seropositive patients: univariate analysis of patient characteristics.* Ten patients with HCMV gB DNA positive tumors but no matching serum available were assumed to be seropositive for HCMV and included in the IgG+ group. p-values were calculated by Chi-square analysis or the Fisher’s exact test if n <5 in a category. p-value was considered significant if < 0.1 and these values are bolded. ER: estrogen receptor; gB: glycoprotein B; HER2: human epidermal growth factor receptor 2; PDGFRα: platelet-derived growth factor receptor alpha; PR: progesterone receptor.HCMV IgG seronegative and seropositive patients with and without detection of HCMV gB in breast tumors: univariate analysis for patient characteristics.* Ten patients with HCMV gB DNA positive tumors but no matching serum available were assumed to be seropositive for HCMV and included in this analysis. p-values were calculated by Chi-square analysis or the Fisher’s exact test if n < 5 in a category. p-value was considered significant if < 0.1 and these values are bolded. ER: estrogen receptor; gB: glycoprotein B; HER2: human epidermal growth factor receptor 2; PDGFRα: platelet-derived growth factor receptor alpha; PR: progesterone receptor.Multivariate analysis of patient characteristics identified by univariate analysis to determine the odds ratio of being HCMV gB DNA and/or IgG positive.* Ten patients with HCMV gB DNA-positive tumors but no matching serum available were assumed to be seropositive for HCMV and included in the IgG+ group. p-values were calculated by multiple logistic regression analysis. p-value was considered significant if < 0.05 and these values are bolded. OR: odds ratio; CI: confidence interval.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01149.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Long noncoding RNAs (lncRNAs) are important regulators of cell progression or regression in gastric cancer (GC). The lncRNA LOC441461 was downregulated in TNM stage IV GC. The downregulation of LOC441461 promoted the proliferation, cell cycle progression, and metastasis of GC cells in vitro. We confirmed, with predictable targets, that LOC441461 interacts with transcription factors and promotes tumor progression. Our study suggests LOC441461 as a potential biomarker, which could also lead to a therapeutic target in patients with advanced TNM stage gastric cancer.Gastric cancer is a common tumor, with a high mortality rate. The severity of gastric cancer is assessed by TNM staging. Long noncoding RNAs (lncRNAs) play a role in cancer treatment; investigating the clinical significance of novel biomarkers associated with TNM staging, such as lncRNAs, is important. In this study, we investigated the association between the expression of the lncRNA LOC441461 and gastric cancer stage. LOC441461 expression was lower in stage IV than in stages I, II, and III. The depletion of LOC441461 promoted cell proliferation, cell cycle progression, apoptosis, cell motility, and invasiveness. LOC441461 downregulation increased the epithelial-to-mesenchymal transition, as indicated by increased TRAIL signaling and decreased RUNX1 interactions. The interaction of the transcription factors RELA, IRF1, ESR1, AR, POU5F1, TRIM28, and GATA1 with LOC441461 affected the degree of the malignancy of gastric cancer by modulating gene transcription. The present study identified LOC441461 and seven transcription factors as potential biomarkers and therapeutic targets for the treatment of gastric cancer.Gastric cancer is a common malignancy, with a high mortality rate worldwide, and shows substantial incidence in East Asia, Eastern Europe, and South America [1]. The two most important risk factors for gastric cancer are Helicobacter pylori infection, which can induce atrophic gastritis, as well as intestinal metaplasia, and salted food intake [1]. H. pylori infection is the most prevalent, infecting more than 50% of all gastric cancers [2]. Common symptoms of gastric cancer include indigestion, anorexia, weight loss, and abdominal pain [1]. Gastric cancer with the persistence of symptoms before diagnosis can be incurable or in an advanced stage [1]. The diagnosis of gastric cancer is determined using the tumor, node, and metastasis (TNM) classification system, according to the American Joint Committee on Cancer [3]. Clinical staging is determined according to physical examination, biopsy, radiological imaging, and the results of an endoscopy [4,5,6]. In patients who undergo surgical resection, pathological staging can be performed, according to the analysis of surgical specimens [4,5,6]. The clinical and pathological staging of gastric cancer is determined by various combinations of the T, N, and M categories [7,8]. Although only 20% of patients survive more than 5 years after diagnosis, there is no universal standard first-line therapy [2,9]. For early gastric cancer with moderate differentiation or without invasion, upfront surgery is the preferred means of management. Perioperative or postoperative chemotherapy with chemoradiation is recommended in the current guidelines [10]. For the most advanced gastric cancer, a fluoropyrimidine chemotherapy and platinum-based doublet are typically the backbone regimen [9,10].Long noncoding RNAs (lncRNAs) are longer than 200 nucleotides and transcribed from noncoding regions [11]. Although previous studies have focused on the role of mRNAs in gastric cancer, recent studies have highlighted the involvement of lncRNAs [12,13]. Due to this nature, lncRNAs have their own structural characteristics that play various roles in epigenetic regulation, transcriptional regulation, and post-transcriptional regulation [14]. LncRNAs regulate histone modifications at the chromatin level through histone methylation and acetylation, DNA methylation, and transcriptional activation, and function together with various transcription factors and post-transcriptional modifications [12]. The function of lncRNAs as a decoy molecule blocks the function of chromosome-folding proteins or transcription regulators [14]. In addition, lncRNAs associate with the regulation of gene expression through sponging miRNAs [14]. As guide molecules, lncRNAs lead specific molecules to target locations [14]. If the target molecules are transcription factors, lncRNAs have a role in cisregulation [14,15]. Alternatively, other lncRNAs play a role in transregulation with histone methylation [14,16]. LncRNAs also act as a scaffold that can recruit numerous molecules [14]. For example, a well-known scaffold lncRNA is an X-inactive specific transcript (Xist) that suppresses the expression of the X chromosome in females, and as a result promotes the assembly of polycomb complexes 1 and 2 [14,16]. From the perspective of mRNA processing, lncRNAs modulate splicing factors through regulating the phosphorylation of splicing factors, and hijacking them [16,17]. Based on the molecular function of lncRNAs, lncRNAs play different roles in development and disease, and are involved in the progression or regression of cancer, suggesting their potential as novel therapeutic targets [18,19]. For example, the expression of HOX transcript antisense RNA, which recruits chromatin modifiers, is associated with metastasis and a poor prognosis in lung cancer [20]. Metastasis-associated lung adenocarcinoma transcript 1 is a novel therapeutic target for decreasing resistance in prostate cancer [21].In this study, we show that the lncRNA LOC441461 is downregulated in clinical stage IV gastric cancer. The function of LOC441461 has not been studied in detail, except in colorectal cancer, in which LOC441461 expression is associated with cell growth and motility [22]. The role of LOC441461 in modulating various cellular functions was investigated in human gastric cancer cell lines. The downregulation of LOC441461 expression increased the aggressiveness of gastric cancer cells, suggesting that decreased LOC441461 is associated with a poor prognosis.Clinical data, survival data, and RNA-seq data of STAD were downloaded from UCSC XENA (https://xena.ucsc.edu/ accessed on 10 December 2021). A total of 379 gastric cancer cases were analyzed for clinical TNM staging and RNA-seq data. The unit of RNA-seq data was log2(RPKM + 1).Clinical TNM staging was performed for each individual. IG was assessed using the clinical staging of each individual, in addition to gene-level expression data using FSelector package v. 0.31 (a package in R, a language and environment for statistical computing, R Core Team, R Foundation for Statistical Computing, Vienna, Austria) [23,24]. Genes with an IG > 0 were used as the feature selection threshold for the clinical stage.A survival analysis was conducted to investigate the difference in prognoses between groups, which were composed of a combination of the T and M categories. A logrank test was conducted with Survival v. 3.2 (https://cran.r-project.org/web/packages/survival/index.html accessed on 8 January 2022). The Kaplan–Meier plot was visualized with Survminer v 0.4.8 (https://cran.r-project.org/web/packages/survminer/index.html accessed on 8 January 2022).All human gastric cancer cell lines (AGS, SNU16, SNU216, SNU638, MKN45, and MKN74) were purchased from the Korean Cell Line Bank (KCLB; Seoul, Republic of Korea) and cultured in RPMI-1640 (Gibco, Grand Island, NY, USA) supplemented with 10% fetal bovine serum (FBS; Gibco) and 1% penicillin/streptomycin (Gibco). Cells were grown at 37 °C in a 5% CO2 incubator, under humidified conditions.The LOC441461 siRNA oligonucleotide (si-LOC441461; BIONEER, Daejeon, Republic of Korea) and negative control scrambled siRNA (N.C, BIONEER) were transfected into MKN74 and SNU216 cells using Lipofectamine™ RNAiMAX (Invitrogen, Carlsbad, CA, USA) and Opti-MEM (Gibco) at a final concentration of 100 nM. After 24 or 48 h, cells were harvested for subsequent experiments. The oligonucleotide sequences of the siRNAs are listed in Table S1.The cytoplasmic or nuclear location of LOC441461 was assessed in MKN74 and SNU216 cells using a PARIS kit (Thermo Fisher Scientific, Waltham, MA, USA) according to the manufacturer’s protocol.Total RNA was extracted from cells using an RNeasy mini kit (Qiagen, Hilden, Germany). The purity and concentration of total RNA were measured using a NanoDrop 1000 spectrophotometer (ND-1000, Thermo Fisher Scientific). One microgram of RNA was used for cDNA synthesis with an AccuPower CycleScript RT Premix (BIONEER) according to the manufacturer’s instructions. qPCR was performed on a CFX96 Touch real-time PCR detection system (Bio-Rad Laboratories, Hercules, CA, USA) using an iQ™ SYBR Green Supermix (Bio-Rad Laboratories) and the indicated primer sets (Table S2). The following PCR reactions were performed: initial denaturation (95 °C for 10 min) followed by 45 cycles of denaturation (95 °C for 10 s) and annealing (55 °C for 30 s). Data were analyzed using the 2−ΔΔCt method and normalized to GAPDH or U6 as an endogenous internal control.Living cells (5000 living cells/well) transfected with N.C or si-LOC441461 were seeded in 96-well culture plates. After 6 h of transfection (or incubation) at 37 °C in 5% CO2 conditions, the medium was replaced with a fresh culture medium. Cell Counting Kit 8 solution (CCK8; Abcam, Cambridge, UK) was added directly to the cells. Cell growth was determined by measuring the absorbance at 450 nm every 24 h.N.C- or si-LOC441461-transfected MKN74 cells were seeded into 6-well plates at a density of 2500 cells per well. After 2 weeks of incubation at 37 °C in 5% CO2, cells were fixed with 3.7% formaldehyde for 10 min and stained with 0.01% crystal violet solution in 10% ethanol for 20 min. Colony formation was quantified using 33% acetic acid and measured on a VICROR3 Multilabel Plate Reader (PerkinElmer, Waltham, MA, USA) at a wavelength of 590 nm.A total of 1 × 106 cells were harvested and fixed with 70% ethanol added dropwise to the pellet. After 12 h of incubation at −20 °C, cells were stained with PI/RNase Staining Buffer (BD Biosciences, San Jose, CA, USA) for 15 min at RT and detected using flow cytometry (Beckman Coulter, Brea, CA, USA). Cell cycle progression was analyzed using FlowJo v. 10.8.1 (FlowJo LLC, Ashland, OR, USA) with the Watson (Pragmatic) model.N.C- or si-LOC441461-transfected MKN74 and SNU216 cells were seeded into 6-well plates at a density of 3 × 105 cells per well and incubated in the presence of 5-FU (2.5 μg/mL). After 24 h, apoptotic cells were detected with an Annexin V Apoptosis Detection Kit (BD Biosciences), following the manufacturer’s instructions using flow cytometry (Beckman Coulter).Migration and invasion assays were performed using Transwell plates (Corning, NY, USA). N.C- or si-LOC441461-transfected MKN74 and SNU216 cells (7.5 × 104) were seeded into the upper chamber of Transwell plates in a serum-free medium. For invasion assays the upper chamber was precoated with Matrigel Matrix (Corning). After incubation under culture conditions for 24 h, the lower chamber of the Transwell plates was fixed with 10% formaldehyde solution, and cells on the upper chamber were removed with cotton swabs. A 0.01% crystal violet solution was used to stain cells, to detect migration and invasion. After air-drying, the stained solution was eluted using 33% acetic acid, and the absorbance of the eluted solution was measured using a VICROR3 Multilabel Plate Reader (PerkinElmer) at a wavelength of 590 nm.For the wound healing assay, a 24-well culture insert dish (ibid GmbH, Münster, Germany) was used to determine the migration ability of MKN74 cells. N.C- and si-LOC441461-transfected MKN74 cells were seeded onto the culture inserts and incubated for 12 h after a confluent monolayer was formed. The inserts were gently removed, and a complete growth medium was added. Wound closure at 24 and 48 h was photographed under a microscope. The wound area was measured and quantified using ImageJ v. 1.53k (National Institutes of Health, Bethesda, MD, USA).The total RNA of N.C- or si-LOC441461-transfected MKN74 cells was extracted with Trizol (Invitrogen), and the concentration was calculated using Quant-IT RiboGreen (Invitrogen). The integrity of the total RNA was measured on TapeStation RNA ScreenTape (Agilent Technologies, Santa Clara, CA, USA), and an RNA integrity number >7.0 was used for RNA library construction.Total RNA (0.5 µg) was used to construct a library using an Illumina TruSeq Stranded Total RNA Library Prep Gold Kit (Illumina, Inc., San Diego, CA, USA). After removing the rRNA, fragmentation was performed with mRNA under elevated temperatures. First-strand cDNA was synthesized using SuperScript II Reverse Transcriptase (Invitrogen) and random primers. Second-strand cDNA synthesis was performed using DNA polymerase I, RNase H, and dUTP. These cDNA fragments went through an end repair process mediated by the addition of a single ‘A’ base, as well as the ligation of the adapters. The products were purified and enriched by PCR to create the final cDNA library.The libraries were quantified using KAPA Library Quantification Kits for Illumina sequencing platforms according to the qPCR Quantification Protocol Guide (KAPA BIOSYSTEMS, #KK4854) and qualified using the TapeStation D1000 ScreenTape (Agilent Technologies, #5067-5582). Indexed libraries were then submitted to Illumina NovaSeq (Illumina), and paired-end (2 × 100 bp) sequencing was performed by Macrogen Incorporated (Seoul, Republic of Korea). Data have been deposited in the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/ accessed on 8 January 2022), under the accession number GSE193700.Differentially expressed genes between N.C and si-LOC441461 groups were analyzed using the DESeq2 package, which is included in R [25]. Genes with an absolute value of log2 fold change > 1 and a Benjamini–Hochberg-adjusted p-value < 0.05 were included.Reactome pathway (https://reactome.org/ accessed on 18 November 2021) enrichment analysis was performed using ShinyGO software, and significant pathways with an FDR correction < 0.05 were extracted in ShinyGO v. 0.741 [26]. The p-values of the enriched pathways are presented as −log10 (p-value).Molecules interacting with LOC441461 were predicted using RNAinter v. 4.0 [27]. Transcription factor enrichment analysis was performed to identify factors affected by LOC441461 using ChEA (https://www.encodeproject.org/chip-seq/ accessed on 18 November 2021) and ENCODE ChIP-seq databases (https://maayanlab.cloud/chea3/ accessed on 18 November 2021), and the differentially expressed gene set between N.C and LOC441461 knockdown samples using ShinyGO [26]. The common transcription factors were visualized using Venny v. 2.1.0 (https://bioinfogp.cnb.csic.es/tools/venny/ accessed on 20 November 2021).The target genes of each transcription factor were extracted according to cotarget genes from ENCODE ChIP-seq data, based on the target gene database [28] and the ChEA ChIP-X target gene database [29].Hierarchical clustering was performed to examine the differences in the expression of target genes between the si-LOC441461 and N.C groups. The expression level of genes was normalized to transcripts per million. ComplexHeatmap v. 2.8.0, a package of R, was used to visualize the heatmap with annotation group information [30].Significant differences between treatment means were determined using Student’s t-test. All qRT-PCR, cell proliferation, colony formation, cell cycle, apoptosis, migration, invasion, wound healing, and RNA-seq assays were performed in triplicate. Error bars indicate standard deviation. One asterisk (*) denotes p < 0.05; two asterisks (**) denote p < 0.01; three asterisks (***) denote p < 0.001; and four asterisks (****) denote p < 0.0001.The result of IG feature selection between TNM stage I and IV gastric cancer patients using RNA-seq data identified 59 genes with an IG value > 0. The 59 genes were subjected to a t-test comparing TNM stage I and stage IV, and seven statistically significant genes were identified after Bonferroni correction. The gene names, including IG scores and p-values determined via a t-test, are presented in Table 1. The IG value results are presented in Table S3. The p-values of genes with an IG >0 determined by t-test are presented in Table S4.The expression of LOC441461 was assessed in samples from different TNM stages and compared between tumor and normal tissues. LOC441461 expression was lower in stage IV than in stages I, II, and III, whereas its expression was similar between stages I, II, and III (Figure 1A). LOC441461 expression did not differ between gastric cancer and adjacent normal tissues (Figure 1B). As LOC441461 expression differed between stage IV and stage I, II, and III gastric cancer, a logistic regression analysis was performed according to the T, N, and M categories. T4 tumors showed lower LOC441461 expression than T1, T2, and T3 tumors, whereas there was no difference between the different N categories. In the M category, LOC441461 expression was lower in M1 than in M0 tumors (Table 2). Based on the difference in expression of LOC441461 in the T and M categories, a survival analysis was conducted with overall survival (OS) and progression-free interval (PFI). The combination of the T category and the M category was divided into six groups. The difference in the prognosis in these six groups was significant in both OS and PFI (Figure 1C,D). The prognoses of T4 with M0 and M1 with T4 were significantly worse than T1 with M0 and M0 with T4, respectively, in OS (Table S5). Otherwise, only T4 with M0 showed a worse prognosis than T1 with M0 in PFI (Table S5). The entire results of the logrank test are described in Table S5.The effect of LOC441461 on gastric cancer cells was investigated based on the TCGA STAD data. LOC441461 expression was first measured in different gastric cancer cell lines, to select the most adequate cells. The MKN74 and SNU216 cell lines showed high expression of LOC441461 and were, thus, selected for further experiments (Figure 2A). MKN74 and SNU216 cells were transfected with si-LOC441461, and depletion of LOC441461 was detected (Figure 2B). LOC441461 expression was higher in the nucleus than in the cytoplasm in both the SNU216 and MKN74 cell lines (Figure 2C).Lower LOC441461 expression was correlated with increased tumor volume, suggesting that negative LOC441461 expression affects the proliferation ability of MKN74 and SNU216 cells. The results of the CCK8 assay showed that LOC441461 significantly inhibited the proliferation of MKN74 and SNU216 cells, suggesting that LOC441461 promotes gastric cancer cell proliferation in vitro (Figure 3A and Figure S1). LOC441461 knockdown induced cell cycle progression; specifically, the G1/S transition was accelerated (Figure 3B). However, the colony formation ability, which indicates the ability of a single cell to grow into a colony, was significantly decreased in response to si-LOC441461 in both MKN74 and SNU216 cells (Figure 3C). The present results indicated that LOC441461 knockdown induced gastric cancer cell apoptosis in vitro. As shown in Figure 4, LOC441461 knockdown promoted apoptosis in MKN74 and SNU216 cells treated with 5-FU (Figure 4A,B). We, thus, examined the effect of LOC441461 on apoptosis and drug resistance in MKN74 and SNU216 cells treated with 5-FU. The results showed that the expression of LOC441461 correlated negatively with the drug sensitivity of gastric cancer (Figure 4A); whereas, DMSO treatment decreased the rate of apoptosis of LOC441461 knockdown cells (Figure 4A,B).Analysis of the migration and invasion of gastric cancer cells using the Transwell assay showed that SNU216 cells had a higher rate of migration and invasiveness than MKN74 cells (Figure S2). The knockdown of LOC441461 increased the migration and invasion capability of SNU216 cells (Figure 5A). Additionally, the migration and invasion abilities of SNU216 cells were also significantly increased in the absence of LOC441461 (Figure 5B). The results of the wound healing assay showed that wound closure was significantly higher in LOC441461-silenced MKN74 cells than in the negative control group (Figure 5C). The decrease in the wound area was statistically significant after 48 h, but not at 24 h.The transcriptomic landscape of LOC441461 knockdown gastric cancer cells was analyzed by RNA sequencing. Three replicates were, respectively, included in two groups, including negative controls and the knockdown of LOC441461. Differentially expressed gene analysis was performed using expression data obtained by RNA-seq. A total of 2503 genes with an absolute log2 fold change >1 and an adjusted p < 0.05 were selected for pathway enrichment analysis. Of these, 1202 upregulated genes and 1301 downregulated genes in the knockdown group were selected. The genes are listed in Table S6. The selected genes were subjected to pathway enrichment analysis using the Reactome pathway database. Apoptosis- and cell-cycle-progression-associated terms were significantly enriched in genes that were upregulated in the LOC441461 knockdown group (Figure 6A). The detailed results are presented in Table S7. The mRNA expression of cyclin D1, which is involved in the G1/S transition of the cell cycle, and TRAIL, which is associated with apoptosis and the epithelial-to-mesenchymal transition (EMT), were higher in the LOC441461 knockdown group (Figure 6B).Analysis of the RNAinter database showed that most of the molecules interacting with LOC441461 were transcription factors (Table S8). To identify the transcription factors modulated by LOC441461, a transcription factor enrichment analysis was performed according to the ENCODE and ChEA consensus transcription factors obtained from the ChIP-seq database, as well as the differentially expressed genes from RNA-seq data (Table S9). RELA and IRF1 interacted with LOC441461 and were significantly enriched among the genes upregulated by LOC441461 knockdown, whereas ESR1, AR, POU5F1, YY1, FOXM1, TRIM28, SMAD4, E2F4, and GATA1 interacted with LOC441461 and were significantly enriched in the downregulated gene set (Figure 7A). Hierarchical clustering was performed based on the target gene expression of significantly enriched transcription factors from RNA-seq data. Transcription factors significantly affected by LOC441461 were identified by hierarchical clustering, which showed that the expression of the RELA, IRF1, ESR1, AR, POU5F1, TRIM28, and GATA1 target genes differed significantly between the knockdown and negative control groups (Figure 7B). The detailed results of hierarchical clustering and the expression of each target gene are presented in Figures S3 and S4. Target genes were extracted by the cotarget genes in the ENCODE ChIP-seq database and ChEA ChIP-X database, except AR and SMAD4, which were extracted from the ChEA ChIP-X database. The target genes of each transcription factor are listed in Table S10.Recent reports on the different functions of lncRNAs have highlighted their involvement in tumor development and progression [18,31]. Moreover, the investigation of the molecular and cellular functions of lncRNAs may lead to the identification of therapeutic targets and the design of strategies for the treatment of gastric cancer [18,31,32,33]. By examining the association between lncRNA LOC441461 expression level and TNM stage, we identified biological processes in gastric cancer. In this study, LOC441461 expression was lower in stage IV than in stage I, II, and III gastric cancer samples from TCGA STAD data. The association between the degree of the malignancy of gastric cancer and higher stages, as assessed by the TNM staging system, supports the effect of LOC441461 downregulation on increasing the severity of the disease by promoting tumor growth, cell motility, and worse prognoses. Furthermore, the localization and local interactions of lncRNAs are key to predicting their function [34]. Overall, we suggest that LOC441461 modulates gene transcription, inducing the malignancy of gastric cancer by interacting with RELA, IRF1, ESR1, AR, POU5F1, TRIM28, and GATA1.LncRNAs are involved in diverse biological processes, such as proliferation, motility, and the EMT in different aspects of regulation [11,35]. The latest research shows that LOC441461 upregulation promotes growth and motility in colon cancer cell lines [22]. In the present study we found that the different effects of LOC441461 may be attributed to the use of gastric cancer, and several experimental validations were conducted to investigate the function of LOC441461 in gastric cancer cells. The results of cell experiments and RNA-seq analysis showed that the downregulation of the lncRNA LOC441461 could promote the growth and metastasis of gastric cancer cells in vitro and in silico. Furthermore, lncRNAs modulate the cell cycle by regulating the expression of cyclins and cyclin-dependent kinases [36,37], and a similar mechanism may underlie the regulation of the cell cycle by LOC441461. The FOXO-mediated transcription of cell cycle genes was significantly increased in LOC441461 knockdown cells. However, the clonogenic ability showed a positive correlation with the expression of LOC441461.5-FU treatment, a fluoropyrimidine chemotherapy, is the first treatment of choice of gastric cancer [9]. Previous studies suggest that 5-FU sensitivity increases the status of DNA damage [38,39]. Ultimately, the p53-independent DNA damage response was significantly enriched in LOC441461 knockdown cells, suggesting that the DNA damage response was promoted by the effect of LOC441461 knockdown, in inducing the 5-FU-induced apoptosis of gastric cancer cells.In addition, it was found that lncRNAs regulate gastric cancer cells via diverse contexts, such as regulating the EMT process and PI3K/AKT pathway [22,31,40]. The EMT is considered as a key role in human malignancies activated during tumor metastasis [41]. Given the abovementioned findings, the results suggested that the knockdown of LOC441461 increased the motility of gastric cancer cells. The downregulation of LOC441461 increased the proportion of migrating/invading cells and promoted wound closure. The upregulation of tumor-necrosis-factor-related apoptosis-inducing ligand (TRAIL) signaling also promotes the EMT in various cancers [42,43,44,45]. By contrast, the interaction of RUNX1 with cofactors was significantly decreased in the absence of LOC441461, and RUNX1 is a suppressor of the EMT [46,47]. Thus, the EMT was promoted in relation to the depletion of LOC441461, as indicated by TRAIL signaling and RUNX1 interactions.Multiple studies have observed that lncRNAs are, overall, more numerous in the nucleus than their cytoplasm [16,31,34,35,48]. Interestingly, the instability of nuclear lncRNAs can act as an oncogene or tumor suppressor and modulate transcription factors, which are key players in transcriptional regulation [16,31,48]. We investigated the potential mechanism underlying the modulation of gene expression by transcription factors that interact with LOC441461 in gastric cancer cells. Transcription factor enrichment analysis was performed to identify transcription factors that interact with LOC441461. Hierarchical clustering based on the expression of the target genes of 11 transcription factors affected by LOC441461 resulted in the classification of the target genes of seven transcription factors, RELA, IRF1, ESR1, AR, POU5F1, TRIM28, and GATA1, into two groups (Figures S3 and S4). Hierarchical clustering of the target genes of seven transcription factors also clearly divided them into two groups, in correlation with the expression of LOC441461. RELA, also called p65, is associated with NF-κB heterodimer formation, nuclear translocation, and activation [49]. The dysregulation of NF-κB/RELA promotes distant metastasis in gastric cancer, suggesting its potential as a novel therapeutic target [50]. IRF1 not only acts as a viral infection response transcription factor [51], but also plays a role in cell cycle arrest, enhancing 5-FU sensitivity and suppressing the EMT in gastric cancer [52,53,54]. In the absence of LOC441461, IRF1 expression was upregulated (Figure S5). ESR1 and androgen receptor (AR) upregulation is associated with a poor prognosis in gastric cancer patients [55]. POU5F1, a POU class 5 homeobox 1 member, is associated with a poor prognosis, and its expression is significantly increased after chemotherapy in colorectal cancer [56,57]. Tripartite motif-containing 28 (TRIM28) is associated with a poor prognosis by promoting tumor progression and the activation of autophagy in glioma [58,59]. GATA1 or GATA-binding factor 1 promote the EMT in breast cancer [60].According to the results of an expression analysis between TNM stages and a survival analysis, the expression of LOC441461 was associated with malignant characteristics of gastric cancer cells, including cell growth and motility. Specifically, the molecular function of LOC441461 was bioinformatically predicted by modulating the activity of seven transcription factors, including RELA, IRF1, ESR1, AR, POU5F1, TRIM28, and GATA1. Based on these facts, they can be applied to the novel treatment management of gastric cancer in various ways. First, upregulating the expression of LOC441461 in gastric cancer patients can suppress progression to a malignant status of gastric cancer. Second, modulating the activity of transcription factors interacting with LOC441461 could be a candidate therapeutic target to manage advanced gastric cancer. Recently, RNA biopharmaceuticals have been attracting attention as a breakthrough clinical application of the COVID-19 RNA vaccine [33,61]. Proving its application as a gastric cancer treatment by regulating the expression of LOC441461 will be a very interesting study in the future.In this study, we identified the lncRNA LOC441461 as a potential biomarker in patients with an advanced TNM stage, which has significant value in tumor development and progression. LOC441461 knockdown promoted growth, cell cycle progression, motility, and invasiveness in gastric cancer cells in vitro. Additionally, molecular links between the expression of LOC441461 and the nature of gastric cancer cells were investigated by measuring changes in gene expression modulated by transcription factors. Candidate transcription factors that may interact with LOC441461 were identified using bioinformatic methods and various databases. Although this study has the limitation that the exact mechanism of action with which LOC441461 interacted with seven transcription factors was only predicted by bioinformatical methods, and not validated by experimental methods, the findings in this study may lead to the development of novel therapeutic strategies for patients with gastric cancer.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051149/s1, Figure S1. Knockdown of LOC441461 by transfection with si-LOC441461 or a scrambled negative control (N.C) in SNU216 cells. The proliferation ability of N.C- and si-LOC441461-transfected cells was examined using the CCK-8 solution at every 24 h. All experiments were performed in triplicate, and data are expressed as the mean ± standard deviation. Figure S2. MKN74 or SNU216 gastric cancer cells were seeded in Transwell plates to evaluate migration and invasion. (A) The results of migration and invasion assays with MKN74 and SNU216 cells stained with 0.01% crystal violet solution. Scale bar = 200 μm. (B) Cell movement was quantified by elution with 33% acetic acid and measuring the absorbance at a wavelength of 590 nm using a plate reader. All experiments were performed in triplicate, and data were expressed as the mean ± standard deviation. Figure S3. Heatmap of the expression of the target genes of transcription factors dividing genes into LOC441461 knockdown and negative control groups. Figure S4. Heatmap of the expression of the target genes of transcription factors that did not divide genes into LOC441461 knockdown and negative control groups. Figure S5. IRF1 mRNA expression was upregulated in the LOC441461 knockdown group. qRT-PCR analysis of MKN74 cells transfected with a scrambled negative control siRNA (N.C) or si-LOC441461. The data were normalized to GAPDH. All experiments were performed in triplicate and expressed as the mean ± standard deviation (*** p < 0.001; Student’s t-test). Table S1. siRNA sequences and target RNA. Table S2. Primer set used in qRT-PCR targeting LOC441461, including TRAIL, cyclin D1, IRF1, U6, and GAPDH. Table S3. Results of IG analysis using TCGA STAD RNA-seq data and TNM stage information from clinical data obtained between stage I and IV. Table S4. Statistical analysis (t-tests) of the expression of genes with an IG >0 at TNM stages I and IV. Table S5. The results of a logrank test. Table S6. Differentially expressed genes between the LOC441461 knockdown and N.C groups. KD = knockdown of LOC441461. N.C = negative controls. Table S7. Results of pathway enrichment analysis using the Reactome database and differentially expressed genes. Blue indicates a downregulated pathway in the LOC441461 knockdown group. Red indicates an upregulated pathway in the LOC441461 knockdown group. Table S8. Molecules that interact with LOC441461. Table S9. Significantly-enriched transcription factors. Table S10. Target genes of each transcription factor.Conceptualization, S.-s.L., J.P., S.O. and K.K.; Data Curation, S.-s.L., J.P. and K.K.; Formal Analysis, S.-s.L., J.P. and S.O.; Funding Acquisition, K.K.; Investigation, S.O. and K.K.; Methodology, S.-s.L. and J.P.; Project Administration, K.K.; Software, J.P.; Supervision, K.K.; Visualization, S.-s.L. and J.P.; Writing—Original Draft, S.-s.L. and J.P.; Writing—Review and Editing, S.O. and K.K. All authors have read and agreed to the published version of the manuscript.This work was supported by the Ministry of Science, Technology, and Information, Republic of Korea (grant no. 2019R1F1A105492013).Not applicable.Not applicable.RNA-seq data have been deposited in the Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/ accessed on 18 November 2021) under the accession number GSE193700 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE193700 accessed on 18 November 2021).Parts of the results are based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga (accessed on 18 November 2021). This study was supported by a Basic Research Project grant from the Ministry of Science, Technology, and Information, Republic of Korea (grant no. 2019R1F1A105492013).The authors declare that they have no competing interests.Comparison of LOC441461 expression and survival analysis using TCGA STAD data. (A) The expression of LOC441461 in each TNM stage and statistical analysis. The statistical analysis was conducted via a t-test. **** p < 0.0001, *** p < 0.001. (B) LOC441461 expression in primary tumors and adjacent normal tissues. ns, not significant. (C,D) Comparison of prognoses between different combinations of T and M. The statistical method was a logrank test. Pathologic_T is the T category of TNM staging. Pathologic_M is the M category of TNM staging.Relative expression of LOC441461 and localization in human gastric cancer cell lines. (A) Expression levels of LOC441461 were measured in six gastric cancer cell lines (MKN74, SNU216, SNU16, MKN45, AGS, and SNU638) by qRT-PCR. (B) Knockdown of LOC441461 in MKN74 and SNU216 cells via transfection with small interfering RNA (siRNA) targeting LOC441461 (si-LOC441461) compared with a scrambled negative control (N.C). qRT-PCR was performed to quantify the relative expression levels of LOC441461 normalized to GAPDH; SNU638 cells were compared as a reference sample. (C) Subcellular fractionation was performed using a PARIS kit to separate the nuclear and cytoplasmic fractions of MKN74 and SNU216 cells. The localization of LOC441461 in the nucleus (normalized to U6) and cytoplasm (normalized to GAPDH) was compared via qRT-PCR using the 2−ΔΔCt method. All experiments were performed in triplicate, and data are expressed as the mean ± standard deviation (** p < 0.01, **** p < 0.0001; Student’s t-test).LOC441461 suppresses cell proliferation by promoting G1/S transition in human gastric cancer cells. (A) Knockdown of LOC441461 by transfection with si-LOC441461 or a scrambled negative control (N.C) in MKN74 cells. The proliferation ability of N.C- and si-LOC441461-transfected cells was examined using the CCK-8 solution at every 24 h. (B) The cell cycle progression of N.C- and si-LOC441461-transfected cells was examined using flow cytometry with the PI/RNase solution. Each cell cycle phase was analyzed with FlowJo v. 10.8.1. Graph of each quantified phase. (C) MKN74 and SNU216 cells were transfected with N.C or si-LOC441461 and stained with 0.01% crystal violet solution after 14 days. (D) Graph of colony formation ability after elution with 33% acetic acid. All experiments were performed in triplicate, and data are expressed as the mean ± standard deviation (* p < 0.05, ** p < 0.01; Student’s t-test).Sensitivity to 5-FU was increased in gastric cancer cells with LOC441461 knockdown. MKN74 and SNU216 cells were transfected with a scrambled negative control (N.C) or si-LOC441461. After exposure to 5-FU (2 μg/mL), apoptotic cells were measured using an annexin V/7-AAD kit. (A) Representative dot plot of the apoptosis assay using flow cytometry. The upper-left (UL) quadrant indicates necrotic cells, the lower-left (LL) quadrant shows viable cells, the lower-right (LR) quadrant indicates early apoptotic cells, and the upper-right (UR) quadrant shows the late stage of apoptosis. (B) Graph showing the percentage of the cells in each stage. Experiments were performed in triplicate, and data are expressed as the mean ± standard deviation (** p < 0.01; *** p < 0.001; Student’s t-test).LOC441461 knockdown promotes gastric cancer cell migration and invasion. (A) Representative microscopic images of the Transwell migration and invasion assays in SNU216 cells (stained with crystal violet). Scale bar = 200 μm. (B) Relative migration and invasion abilities were quantified after elution with 33% acetic acid. (C) Representative images of wound healing assays using culture preinserts in MKN74 cells with LOC441461 knockdown. Scale bar = 500 μm. (D) Wound closure was assessed by ImageJ v. 1.53k from 0 to 48 h. All experiments were performed in triplicate, and data were expressed as the mean ± standard deviation (**** p < 0.0001; Student’s t-test).Transcriptomic landscape after the depletion of LOC441461 in gastric cancer cell lines. (A) Results of pathway enrichment analysis. Red indicates upregulated pathways in LOC441461 knockdown groups. Blue indicates downregulated pathways in LOC441461 knockdown groups. (B) mRNA expression of cyclin D1 and TRAIL determined by qRT-PCR in MKN74 cells transfected with N.C or si-LOC441461. The data were normalized to GAPDH. All experiments were performed in triplicate, and data are expressed as the mean ± standard deviation (**** p < 0.0001; Student’s t-test).Investigation of the role of LOC441461 in modulating transcription factor activity. (A) Enriched transcription factors in the up- or downregulated gene sets in interacting with LOC441461. (B) Hierarchical clustering based on the target gene expression of 11 transcription factors with RNA-seq data of LOC441461 knockdown. KD, knockdown of LOC441461. N.C, negative control.Selected genes with IG and p-values according to t-tests based on XENA TCGA STAD.IG, information gain. p, p-value of t-test.Results of the logistic regression analysis of TNM scores based on XENA TCGA STAD data.CI, confidence interval. p, p-value determined via logistic regression. * Statistically significant p-value from logistic regression analysis after FDR correction.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01150.txt

@@ -0,0 +1,20 @@
+Topic models are algorithms introduced for discovering hidden topics or latent variables in large, unstructured text corpora. Leveraging on analogies between texts and gene expression profiles, these algorithms can be used to find structures in expression data. This work presents an application of topic modeling techniques for the identification of breast cancer subtypes. In particular, we extended a specific class of topic models to allow a multiomics approach. As an illustrative example, considering both messenger RNA and microRNA expression levels, we were able to clearly distinguish healthy from tumor samples as well as the different breast cancer subtypes. The integration of different layers of information is crucial for the observed classification accuracy. Our approach naturally provides the genes and the microRNAs associated to the specific topics that are used for sample organization. We show that indeed these topics often contain genes involved in breast cancer development and are associated to different survival probabilities.The integration of transcriptional data with other layers of information, such as the post-transcriptional regulation mediated by microRNAs, can be crucial to identify the driver genes and the subtypes of complex and heterogeneous diseases such as cancer. This paper presents an approach based on topic modeling to accomplish this integration task. More specifically, we show how an algorithm based on a hierarchical version of stochastic block modeling can be naturally extended to integrate any combination of ’omics data. We test this approach on breast cancer samples from the TCGA database, integrating data on messenger RNA, microRNAs, and copy number variations. We show that the inclusion of the microRNA layer significantly improves the accuracy of subtype classification. Moreover, some of the hidden structures or “topics” that the algorithm extracts actually correspond to genes and microRNAs involved in breast cancer development and are associated to the survival probability.A crucial problem in modern computational biology is the integration of different sources of information in the framework of the so-called “precision medicine” [1]. Thanks to the impressive improvement of experimental techniques and the creation of dedicated databases, plenty of different ’omics datasets are available. However, these datasets are difficult to integrate in a coherent picture. They are typically noisy and sparse; they can strongly depend on experimental and processing choices and biases, such as normalization or imputation techniques, and present different constraints—for example, due to (often unknown) specific regulatory interactions. At the same time, only by combining different layers of information can we hope to understand complex pathologies such as cancer and, thus, optimize the therapeutic protocols. In fact, a major goal would be to be able to identify as soon as possible the particular cancer subtype of a given patient, find the corresponding drivers and altered pathways, and thus, possibly, fine-tune the therapy. A fundamental preliminary step is the development of algorithms able to identify and extract the relevant structure and organization of tumor samples using the different available layers of molecular information.In particular, topic modeling has been recently proposed as a computational technique to identify hidden structures in gene expression data [2,3]. Topic models are a set of algorithms originally developed to extract latent variables from text corpora [4,5,6]. The most popular of these algorithms is the so-called Latent Dirichlet Allocation [5] (LDA), which has been successfully applied not only in texts analysis, but also in other contexts such as bioinformatics [7].LDA is based on the assumption of a Dirichlet prior for the latent variables. This choice simplifies the statistical inference problem making the algorithm highly efficient. However, many complex systems in which LDA is applied, including expression data, are characterized by the emergence of power-law distributions, which are very far from the Dirichlet assumption [8,9,10,11]. Moreover, the optimal number of topics must be identified by the user in the standard LDA formulation [5].To overcome these problems, a new class of algorithms based on hierarchical Stochastic Block Modeling (hSBM) was recently proposed [10]. These algorithms are based on the formal equivalence between the topic identification problem and the community detection problem in bipartite networks [12,13,14], where well-developed techniques based on stochastic block modeling [15] can be applied without the need of a Dirichlet prior.We recently performed a comparative study [3] of different topic modeling algorithms on the task of identifying cancer subtypes from breast and lung cancer gene expression datasets from The Cancer Genome Atlas (TCGA) [16,17]. We found that hSBM typically outperforms other algorithms in the clustering task. Importantly, this algorithm presents the additional advantages of naturally selecting the number of clusters and of providing the genes significantly associated with the latent structure on which the classification is based. We were able to show that the established cancer subtype organization for both breast and lung cancer was well-reconstructed by the latent topic structure inferred by hSBM and that the topic content itself was very informative. In fact, topics associated with specific cancer subtypes were enriched in genes known to play a role in the corresponding disease, and were related to the survival probability of patients.This paper extends our previous study by integrating in the hSBM framework multiple layers of information. While the integration of additional biological information should generally improve the accuracy of the statistical inference, it is important to stress that this is not always trivially true. Highly noisy or irrelevant data layers could interfere with the task. We will show an empirical example of such a negative interference. Therefore, the addition of new layers should be driven by a clear biological motivation.We will focus on the illustrative case of breast cancer, which is the most commonly diagnosed cancer type and the leading cause of cancer death in women worldwide [18], with three main goals:First, we will show how different layers of biological information can be efficiently integrated in the hSBM framework. We release the python package nSBM, inherited from hSBM [10], which is ready to install, easily executable, and can be used to infer the topic structure starting from different layers and types of biological data.Second, focusing on breast cancer, we will show that the combination of microRNA and protein-coding expression levels greatly improves the algorithm’s ability to identify cancer subtypes. These findings further confirm the important role previously recognized in several studies that miRNAs play in cancer development [19,20].Third, we use the inferred topic structure to select a few genes, miRNAs, and chromosomal duplications that seem to have a prognostic role in breast cancer and, thus, could be introduced as additional signatures of specific breast cancer subtypes. The extension of subtype signatures can help clinicians to fine-tune diagnostic protocols in the framework of a precision medicine approach to cancer [1].First, we will show how different layers of biological information can be efficiently integrated in the hSBM framework. We release the python package nSBM, inherited from hSBM [10], which is ready to install, easily executable, and can be used to infer the topic structure starting from different layers and types of biological data.Second, focusing on breast cancer, we will show that the combination of microRNA and protein-coding expression levels greatly improves the algorithm’s ability to identify cancer subtypes. These findings further confirm the important role previously recognized in several studies that miRNAs play in cancer development [19,20].Third, we use the inferred topic structure to select a few genes, miRNAs, and chromosomal duplications that seem to have a prognostic role in breast cancer and, thus, could be introduced as additional signatures of specific breast cancer subtypes. The extension of subtype signatures can help clinicians to fine-tune diagnostic protocols in the framework of a precision medicine approach to cancer [1].Many real-word networks are accompanied by annotations or metadata describing different node properties. For example, in social networks, information about age, gender, or ethnicity can be associated to the nodes or the data capacity can be associated to the nodes of the Internet network [21]. In a similar way, different ’omics can provide additional information to biological networks. These metadata can improve the performance of community detection algorithms by providing additional levels of node correlations that are not accessible only using a single data source [22,23,24]. Given the relation between community detection and topic modeling [10], a similar improvement is expected also in the detection of latent variables using topic modeling analysis on multiomics datasets. Our first goal is, thus, to extend the topic modeling approach to multiomics data, and to test its performances in a concrete biological problem.The extension of a network-based topic modeling algorithm to multipartite networks was recently proposed in the classic context of text analysis by [23], and we apply here a similar approach to biological data. In this case, networks are generic n-partite networks that contain nodes of n types: sample nodes (i.e., patients), and (n−1) sets of nodes (e.g., protein-coding mRNA levels, microRNAs) that represent different features associated with the sample nodes.The topology of the n-partite network is starlike with a center containing the sample nodes and n−1 branches (Figure 1b). Each node in a branch can be connected with all the sample nodes, but no connection exists between nodes within a branch nor between nodes in different branches. This is the natural generalization of the standard bipartite network shown in Figure 1a. In the biological example that will be addressed in the following, only two branches are present: protein-coding genes and microRNAs. However, the presented scheme is general and can be easily extended to several branches at the expense of computational speed. We will discuss the addition of a third sample feature capturing the gene Copy Number Variation (CNV).We shall denote in the following as “links” the connections between the branch nodes and the sample nodes. Each link is characterized by a weight. The weights can have a different nature depending on the branch. For instance, weights on links connecting the gene branch with the samples encode the expression level (here in FPKM units) and, analogously, the links connecting to miRNAs report the miRNA expression level. When we add a layer with the CNV information, the links are weighted with the number of copies of the gene in the connected sample. The algorithm interprets the weight wij between node i and node j as a collection of wij independent edges. We will use the term “edge” for this elementary unit of link weights.Once the multipartite network is defined, the statistical inference procedure leading to the topic structure is a straightforward extension of the procedure developed for the hierarchical Stochastic Block Model (hSBM) [10], which we already applied in its bipartite form to expression data [25]. hSBM is a generative model that basically searches the parameters (θ) that maximize the probability that the model describes the data (A)
+        P(θ|A)∝P(A|θ)P(θ).The model uses a generative process to build a network given a set of parameters θ. Using a Markov Chain Monte Carlo algorithm, these parameters are optimized in a unsupervised way and the optimization continues until the generated model approximates well the data A. (see [10] and references therein for more details).The output of the algorithm is a partitioning of nodes or a set of “blocks” of nodes associated to probability distributions. The samples are partitioned into “clusters”, while the blocks of nodes in the branches are essentially the “topics”. Since we are considering several branches, we will have topics of different types, such as gene-topics on the gene expression branch, miRNA-topics on the miRNA branch, CNV-topics on the CNV branch, and so on. We will consider clusters and topics as “hard” blocks (i.e., each sample/gene/miRNA belongs to only one block) and distinct (there are no blocks containing different kind of nodes). However, given its probabilistic nature, the algorithm can be naturally extended to fuzzy clusters.There are several features that distinguish hSBM, and its nSBM extension introduced here, from other clustering or topic modeling algorithms such as LDA.
+        Lack of a parametric prior.Thanks to the network-based approach and to the particular way links are used to update the block structure, this class of algorithms does not require a specific parametric assumption for the prior probability distribution of the latent variables. This is a major difference with respect to LDA and makes this class of algorithms particularly suited for biological systems in which long-tail distributions and hierarchical structures are ubiquitous (see the discussion on this point and the comparison with LDA in [3]).Probability distributions over latent variables of different types.The output of the algorithm is not deterministic but is instead a set of probabilities that associate a sample with latent variables of different types P(gene-topic|sample), P(miRNA-topic|sample) and associate different features to topics, such as P(gene|gene-topic) and P(miRNA|miRNA-topic). P(gene-topic|sample) and P(miRNA-topic|sample) represent the contribution of each miRNA- or gene-topic to each sample. On the other hand, P(gene|gene-topic) and P(miRNA|miRNA-topic) quantify how much each gene or miRNA contributes to a specific topic.As we will show in the following, these probability distributions capture relevant properties of the biological system.Hierarchical topic structure.Blocks and the probability distributions described above are available at different layers of resolution, from few large sets (clusters/gene-topics/miRNA-topics) at low resolution to many small sets at a higher resolution. The specific number of layers and their block composition are found by the algorithm optimization process and are not given as input. Therefore, the datasets can be organized in different ways depending on the resolution of interest. Note that not all possible resolutions are trivially present, as in standard hierarchical clustering.Concurrent and separate topic organization of the different network layers.Different ’omics have typically different normalization, and the numbers associated to different molecular features have often a very different meaning. A major advantage of nSBM with respect to other algorithms [26,27] is that each layer is independently contributing to the optimization process and a topic organization is given for each layer. Therefore, there is no need to reweight the different layers to balance their contributions since they are kept separate while concurrently contributing to the sample clustering. This makes the model suitable to be applied not only to genomics data, as we will discuss in this paper, but, ideally, to any combination and number of different concurrent ’omics.Lack of a parametric prior.Thanks to the network-based approach and to the particular way links are used to update the block structure, this class of algorithms does not require a specific parametric assumption for the prior probability distribution of the latent variables. This is a major difference with respect to LDA and makes this class of algorithms particularly suited for biological systems in which long-tail distributions and hierarchical structures are ubiquitous (see the discussion on this point and the comparison with LDA in [3]).Probability distributions over latent variables of different types.The output of the algorithm is not deterministic but is instead a set of probabilities that associate a sample with latent variables of different types P(gene-topic|sample), P(miRNA-topic|sample) and associate different features to topics, such as P(gene|gene-topic) and P(miRNA|miRNA-topic). P(gene-topic|sample) and P(miRNA-topic|sample) represent the contribution of each miRNA- or gene-topic to each sample. On the other hand, P(gene|gene-topic) and P(miRNA|miRNA-topic) quantify how much each gene or miRNA contributes to a specific topic.As we will show in the following, these probability distributions capture relevant properties of the biological system.Hierarchical topic structure.Blocks and the probability distributions described above are available at different layers of resolution, from few large sets (clusters/gene-topics/miRNA-topics) at low resolution to many small sets at a higher resolution. The specific number of layers and their block composition are found by the algorithm optimization process and are not given as input. Therefore, the datasets can be organized in different ways depending on the resolution of interest. Note that not all possible resolutions are trivially present, as in standard hierarchical clustering.Concurrent and separate topic organization of the different network layers.Different ’omics have typically different normalization, and the numbers associated to different molecular features have often a very different meaning. A major advantage of nSBM with respect to other algorithms [26,27] is that each layer is independently contributing to the optimization process and a topic organization is given for each layer. Therefore, there is no need to reweight the different layers to balance their contributions since they are kept separate while concurrently contributing to the sample clustering. This makes the model suitable to be applied not only to genomics data, as we will discuss in this paper, but, ideally, to any combination and number of different concurrent ’omics.The benchmark task we now focus on to test the performance of nSBM is its ability to cluster breast cancer samples according to their subtype annotation. This is an important task for its clinical relevance, but also because the breast cancer subtype could be dependent on a complex combination of factors, including gene and miRNA expression profiles; thus, the classification could be a good test for nSBM.Breast cancer is indeed a heterogeneous disease, with wide variations in tumor morphology, molecular characteristics, and clinical response [18,28,29,30]. Notwithstanding this variability, it is one of the few tumors for which there is a widely accepted subtype classification [28,31].Breast cancer samples are usually divided into five different subtypes: Luminal A, Luminal B, Triple-Negative/Basal, HER2, and Normal-like. For our tests, we used as a benchmark the TCGABiolinks annotations [32,33], as discussed in the Methods section. These annotations are the result of a rather complex process. On the clinical side, the classification is based on the levels of a few proteins whose presence in the biopsy are usually detected using immunohistochemistry (IHC) assays. In particular, these proteins are two hormone-receptors (estrogen-receptor (ER) and progesterone-receptor (PR)); the Human Epidermal growth factor Receptor 2 (HER2); and Ki-67, which is a nuclear antigen typically expressed by proliferating cells and, thus, is used as an indicator of cancer cell growth. On the gene expression side, the same subtypes can be identified by looking at the expression levels of a set of genes included in the so-called “Prediction Analysis of Microarray (PAM)50” [34]. The agreement between PAM50 results and IHC-based subtyping is, in general, reasonably good but far from being perfect. Indeed, the classification task is made particularly difficult by the heterogeneity of cancer tissues (biopsies may contain relevant portions of healthy tissue) and by the intrinsic variability of gene expression patterns in cancer cell lines.We recently demonstrated that topic-modeling-based algorithms can achieve satisfactory performances in this classification task by looking at gene expression profiles [3] (and not only of the PAM50 genes), and not relying on the known IHC markers. The advantage of this approach is that it avoids problems and ambiguities in classification due to the stochastic fluctuations of the IHC markers or due to the different inference strategies adopted by PAM50 classifiers (see, for instance, [35] for a recent comparison of the performances of different classifiers in a set of breast cancer classification tasks).Following this line, one of the goals of our study is to evaluate if the integration of miRNA expression levels (and possibly of other layers of information) can further improve the hSBM results presented in ref. [3].It is, by now, well-established that miRNAs play an important role in several human diseases, particularly in cancer. Accordingly, miRNAs have been proposed as diagnostic biomarkers of human cancers [20,36,37]. This is particularly true for breast cancer, for which several studies have highlighted the prognostic role of miRNAs [38].Following this line of evidence, we integrated miRNA expression levels with protein-coding mRNA levels using a n=3 version of nSBM (which, in the following, we shall denote as triSBM). In this case, the analysis output, besides the clusters of samples and the topics of genes, will also contain a collection of miRNA-topics.We first tested if the integration of miRNAs has an effect on the partition of samples in clusters and on the topic organization in the gene branch.Figure 2 reports the Adjusted Mutual Information (AMI) between the partition obtained with a standard hSBM and with triSBM while varying the hierarchy level (l0,l1…), with l0 being the finer layer (the one with smaller sets). We used the AMI to score the overlaps of partitions, since it measures the mutual information between partitions compared with the one obtained by two random partitions. Figure 2a shows that there is a substantial disagreement between the clusters of samples in the two outputs. Similarly, Figure 2b indicates that the same is true for the topics on the protein-coding gene side. The overlap between the partitions obtained by hSBM and triSBM is negligible.Therefore, the addition of the miRNA branch can radically affect the inferred topic structure and the clustering of samples.We first tested the ability of the algorithm in recognizing healthy from cancer samples. The hSBM algorithm showed good performances on this task by considering only gene expression data [3], as summarized in Figure 3b. We then tested triSBM, in which gene expression levels were considered jointly with miRNA levels in the same set of TCGA samples. The detailed procedure and the algorithm output at different hierarchical levels are described in the Methods section. We found a significant improvement in the performance of the algorithm. In fact, Figure 3a clearly shows that normal samples are collected in a single cluster by triSBM, while the separation is less neat in the absence of information on miRNA expression (Figure 3b).The two model settings (hSBM and triSBM) are compared quantitatively in Figure 4 using Normalized Mutual Information (NMI) as a score [39,40]. The NMI score is explained in detail in the Methods section.As a second benchmark, we tested the ability of triSBM to identify breast cancer subtypes. Again, triSBM and hSBM are compared and the results are reported in Figure 5. Further, in this case, the inclusion of miRNA levels improves the algorithm ability to group samples belonging to the same cancer subtype. The improvement is quantified by the NMI scores reported in Figure 5a–c, which show that the improvement is mainly due to the better performance of triSBM in distinguishing LuminalA from LuminalB samples. This was indeed the critical obstacle limiting the performances of hSBM in our previous study [3], suggesting that the distinction of these subtypes crucially depends on miRNA expression levels.We used the Subtype Selected labels provided by TCGABiolinks [32,33] as the ground-truth annotation of subtypes. However, note that this labeling has a less-solid basis with respect to the clear healthy/cancer distinction since the subtypes may not be so clearly defined and can be easily misclassified because of the high tumor heterogeneity.Note that the standard characterization of breast cancer subtypes relies on the expression level of only few markers. We did not explicitly select these markers in our gene selection process; thus, as previously discussed [3], the emergent sample organization is the result of the global pattern of gene and miRNA expression levels. Therefore, the significant overlap with the standard subtype annotation is highly nontrivial, and the discrepancy does not have to be automatically interpreted as a failure since the standard annotation could be limited.Given these positive results, we will explore in the following sections the biological information contained in the latent variables inferred by the algorithm and test their possible prognostic role.We compared the blocks we obtained in output with the annotations of TCGA sample in [41]. First of all, we measured the Adjusted Mutual Information (AMI) between these labels and the Subtype_Selected ones discussed above (AMI is a score between 0 and 1, which measures the mutual information between two annotations compared with the one obtained by comparing two random annotations). We found a value of ∼0.37, which shows that the two labels are not trivially the same and, thus, represent a reliable test of our clusters.We measured the Normalized Mutual Information score of both the bipartite (hSBM) model and the model that integrates miRNA (triSBM). Results are reported in Supplementary Figure S1. Looking at the figure, we see that our clusters also show a significant agreement (high values of NMI/NMI∗) with this independent classification and, above all, that the agreement improves when including miRNAs.The overlap between our cluster partition and two independent nonoverlapping labels can be explained by the fact that our partition groups samples at the intersection between the two labeling systems.We applied the same pipeline applied on TCGA to the METABRIC [42] dataset and measured the agreement between our partition on this data and the labels provided by [41]. We confirmed the results obtained on TCGA: the triSBM model has a better agreement (NMI score is reported in Supplementary Figure S2) with the labels assumed as ground truth with respect to the model without miRNA (hSBM).A major advantage of a topic modeling approach to multiomics data is that we can use the information stored in the probability distributions P(topic|sample) to obtain subtype-specific signatures. Following the analysis of [3], we constructed from these probabilities a set of “centered” distributions P¯(topic|subtype) (see the definition in the Equation (5) of the Methods section), which allow us to identify subtype-specific topics (i.e., topics that are particularly enriched in the samples belonging to a particular subtype) that are candidates to play a role in driving the specific features of that subtype.These topics are nothing but lists of genes and can be investigated using a standard enrichment analysis. The results shown in this paper were computed using the Gene Set Enrichment Analysis GSEA [43] tool. In particular, we concentrated on the keywords extracted from [44,45,46].We discuss the results of this analysis in the following two subsections.We report in Figure 6 a few examples of P¯(gene-topic|subtype) distributions for a few selected topics and in Table 1 the results of the corresponding enrichment analysis.Looking at the figures and at the table, we see a few interesting patterns.
+          There are topics, such as, for instance, topic 8 in Figure 6, which shows a similar behavior in all cancer subtypes and a different one (in the case of topic 8, it is depleted) in the normal tissues. These are the topics that allowed the algorithm to distinguish so accurately normal from cancer samples. In the case at hand, the functional analysis allows to easily understand the reason of this different behavior: the genes contained in topic 8 are strongly enriched in cell cycle keywords, which are likely to be associated to the proliferating nature of tumor tissues.Another interesting pattern is well-exemplified by topics 27, 28, and 44 in Figure 6. These are topics that are over-represented only in one particular subtype (in the example, topics 28 and 44 in the Basal subtype and topic 27 in the HER2 one) and can thus be used as signatures of these subtypes. This is in nice agreement with the finding of the gene enrichment analysis, which, for topics 28 and 44, provides a strong enrichment for the keyword SMID_BREAST_CANCER_BASAL_UP, which is known to be associated with the Basal subtype [44], while topic 27 is enriched in the keyword SMID_BREAST_CANCER_ERBB2_UP, which is in fact associated with the HER2 subtype [44]. These topics are the latent variables that allow the algorithm to distinguish among different subtypes.There are topics, such as, for instance, topic 8 in Figure 6, which shows a similar behavior in all cancer subtypes and a different one (in the case of topic 8, it is depleted) in the normal tissues. These are the topics that allowed the algorithm to distinguish so accurately normal from cancer samples. In the case at hand, the functional analysis allows to easily understand the reason of this different behavior: the genes contained in topic 8 are strongly enriched in cell cycle keywords, which are likely to be associated to the proliferating nature of tumor tissues.Another interesting pattern is well-exemplified by topics 27, 28, and 44 in Figure 6. These are topics that are over-represented only in one particular subtype (in the example, topics 28 and 44 in the Basal subtype and topic 27 in the HER2 one) and can thus be used as signatures of these subtypes. This is in nice agreement with the finding of the gene enrichment analysis, which, for topics 28 and 44, provides a strong enrichment for the keyword SMID_BREAST_CANCER_BASAL_UP, which is known to be associated with the Basal subtype [44], while topic 27 is enriched in the keyword SMID_BREAST_CANCER_ERBB2_UP, which is in fact associated with the HER2 subtype [44]. These topics are the latent variables that allow the algorithm to distinguish among different subtypes.While the above results were similar to the ones already discussed in [3], the novelty of the present analysis is that we can perform a similar study also on the miRNA side. As we will see, this allows for a new independent insight on the problem.We report four instances of the P¯(miRNA-topic|subtype) probability distributions in Figure 7 and the corresponding enrichment analysis in Table 2. They are, somehow, paradigmatic examples of the type of information that one can obtain from this type of analysis.
+          The first one (named miRNA-topic 7 in our output, see https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/trisbm/trisbm_level_0_topics.csv, accessed on 10 February 2022) is the typical example of a topic that shows no particular preference for a cancer subtype (see Figure 6) but shows a strong enrichment for a particular chromosomal locus: chr14q32 (see Table 2). This enrichment is due to the fact that most of the miRNAs of the topic are indeed contained in this locus. Moreover, looking at Figure 8, we see that these miRNAs are exactly those with the highest probability to belong to the topic. This strongly suggests that a somatic alteration (duplication or deletion) at this locus could be associated to the onset of cancer and could thus be used as a marker. Indeed, this locus is known to be associated with breast cancer [47]. Accordingly, if we perform a survival analysis between patients with this topic upregulated and patients with the topic downregulated (see next subsection), we find a remarkable increase in the survival probability of patients with the topic downregulated.However, this is not the end of the story. Looking at Table 2, we see that the same topic is also enriched in keywords associated to Alzheimer disease. Indeed, it is known that there is a sort of inverse comorbidity [48] between a few types of cancer (in particular, lung [49] and breast [50]) and Alzheimer’s disease. This association is confirmed and supported by our analysis, which also suggests that it could be mediated exactly by the microRNAs contained in miRNA-topic 7. Indeed, some of the miRNAs contained in the topic, such as mir-34c, are known oncosuppressors of breast cancer [51,52] and, at the same time, are recognized markers of Alzheimer’s disease [53,54]. The most important of these is the abovementioned mir-34c, which is in fact, strongly associated with miRNA-topic 7, being the only miRNA in the topic with P(miRNA|miRNA-topic)>0.04 not belonging to the locus chr14q32 (see Figure 8).A second class of topics is represented by the other three entries of Figure 7 (miRNA-topics 11, 13, and 16 in our output), which show a different behavior in one of the subtypes with respect to the others (in the present case, these topics are upregulated in samples belonging to the basal subtype). Out of these, only miRNA-topic 11 shows a significant entry in the table of enriched keywords: Table 2. The enrichment is for another chromosomal locus: chr19q13. What is interesting is that this locus has been associated in the past to other types of cancer [55]. Our analysis suggests that it could also play a role in breast cancer and, in particular, in the Basal subtype.The first one (named miRNA-topic 7 in our output, see https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/trisbm/trisbm_level_0_topics.csv, accessed on 10 February 2022) is the typical example of a topic that shows no particular preference for a cancer subtype (see Figure 6) but shows a strong enrichment for a particular chromosomal locus: chr14q32 (see Table 2). This enrichment is due to the fact that most of the miRNAs of the topic are indeed contained in this locus. Moreover, looking at Figure 8, we see that these miRNAs are exactly those with the highest probability to belong to the topic. This strongly suggests that a somatic alteration (duplication or deletion) at this locus could be associated to the onset of cancer and could thus be used as a marker. Indeed, this locus is known to be associated with breast cancer [47]. Accordingly, if we perform a survival analysis between patients with this topic upregulated and patients with the topic downregulated (see next subsection), we find a remarkable increase in the survival probability of patients with the topic downregulated.However, this is not the end of the story. Looking at Table 2, we see that the same topic is also enriched in keywords associated to Alzheimer disease. Indeed, it is known that there is a sort of inverse comorbidity [48] between a few types of cancer (in particular, lung [49] and breast [50]) and Alzheimer’s disease. This association is confirmed and supported by our analysis, which also suggests that it could be mediated exactly by the microRNAs contained in miRNA-topic 7. Indeed, some of the miRNAs contained in the topic, such as mir-34c, are known oncosuppressors of breast cancer [51,52] and, at the same time, are recognized markers of Alzheimer’s disease [53,54]. The most important of these is the abovementioned mir-34c, which is in fact, strongly associated with miRNA-topic 7, being the only miRNA in the topic with P(miRNA|miRNA-topic)>0.04 not belonging to the locus chr14q32 (see Figure 8).A second class of topics is represented by the other three entries of Figure 7 (miRNA-topics 11, 13, and 16 in our output), which show a different behavior in one of the subtypes with respect to the others (in the present case, these topics are upregulated in samples belonging to the basal subtype). Out of these, only miRNA-topic 11 shows a significant entry in the table of enriched keywords: Table 2. The enrichment is for another chromosomal locus: chr19q13. What is interesting is that this locus has been associated in the past to other types of cancer [55]. Our analysis suggests that it could also play a role in breast cancer and, in particular, in the Basal subtype.Moreover, we found a nontrivial overlap between genes in these miRNA-topics and the miRNA clusters proposed by [56]. In particular, there were 12 miRNAs in miRNA-topic 7 from cluster cl349_chr14 (estimating the probability of this happening by chance using a hypergeometric test, we obtained a P-value≃10−5.8), and 8 miRNAs in miRNA-topic 11 were assigned with label cl590_chr19 (P-value≃10−7.4).In the next subsection, we shall study in detail—as an example of the type of analyses that we can perform using the probability distributions obtained from triSBM—the first of these topics.We can use the information contained in the probability distribution P(miRNA|miRNA-topic) to perform a more refined analysis of the miRNAs contained in the topic. First, we see that 75% of the miRNAs in the topic are annotated with the chr14q32 locus and that they are exactly those with the highest values of P(miRNA|miRNA-topic). This can be visualized in Figure 8, where we highlighted in red the miRNAs annotated to the chr14q32 keyword from GSEA [43].Then, we sorted the miRNAs on the basis of their value of P(miRNA|miRNAs’topic) and investigated the first ones (see those with P(miRNA|miRNAs’topic)>0.030 in Table 3); it turns out, using the DISEASES tool [57], that most of them are in some way associated with breast cancer. Let us highlight that mir-511, mir-31, and mir-34c are highly important in this miRNA-topic; nevertheless, they do not belong to the c14q32 locus gene set. What is interesting in our analysis is it suggests that these miRNAs, which were studied in the past as separated entities, are most-probably working together. A better understanding of this cooperative behavior could be of great importance to fine-tune future therapeutic protocols. As a first step in this direction, we took advantage of the probabilistic nature of topic modeling to investigate the survival probability of patients.In particular, since a P(topic|sample) can be assigned to each patient (sample), it is possible to create cohorts of patients based on the importance of a given topic in their transcriptome.We ran a Cox [58] model to verify which is the contribution of our topic to the survival probability of patients.We report in Figure 9 the Kaplan–Meyer curves that we obtained.The contribution of the topic to the survival probability turns out to be very significant: a positive regulation corresponds to higher hazard ratios, meaning that if miRNA inside our topics are expressed higher than normal, the survival probability of patients decreases. While these results should be taken with some caution due to the several sources of bias that may be present in the TCGA population that we tested, it is nevertheless interesting to notice that the presence or absence of this topic has an impact on the survival probability larger than the tumor stage, which is, obviously, strongly correlated with the patient’s prognosis (see Supplementary Figure S3). As a comparison, we also report in Figure S3 variables such as gender (this is not very balanced, as samples are almost all females) or age, which, as expected, do not have significant effects on the survival probability of patients.Going further in the investigation of the survival probability of the patient, one can wonder if patients in a cluster share a similar prognosis.If one measures the fraction of patients still alive 3 years after the diagnosis, it is possible to give a prognosis indication of patients in a given cluster. In Figure S7, we reported two clusters in which the prognosis of the patient is significant. In cluster 6, for instance, only 18% of the patients survived more than 3 years. This corresponds to a cluster with a bad prognosis. On the opposite side, more than 60% of patients grouped in cluster 14 survived: we can assert that patients in this set have a favorable prognosis. We measured the significance of these results by comparing the aforementioned percentages to the ones obtained by creating clusters at random (picking up patients from the whole dataset at random 100 times) and obtained significant Z∼3 scores (reported in Figure S7).There are two main directions in which the analysis discussed in the previous section could be improved. First, one could include in the investigation the regulatory interactions among miRNAs and target genes. Second, one would like to extend the integration to other information layers. We shall discuss in this section a few preliminary attempts in these directions.MiRNAs exert their biological function by regulating target genes at the post-transcriptional level. It is thus of great importance to be able to include this information in the topic modeling analysis. This is not an easy task, since miRNAs act in a combinatorial way: typically, several miRNAs cooperate to regulate a single target gene; at the same time, a single miRNA can regulate hundreds of targets. Moreover, while the standard miRNA–target regulatory interaction is of inhibitory type, it sometimes happens that a miRNA can have a widespread (indirect) activatory role by interfering with a repressed epigenetic pathway. These are the so-called “epi-miRNAs” [59,60] that have been recently shown to play an important role in cancer development [60]. Keeping track of these interactions can be of crucial importance to correctly decode the information contained in the miRNA expression data. To this end, one can make use of a few specialized databases of miRNA–target interactions. In particular, in the following, we shall use MirDip [61] and TarBase [62], which are among the most popular ones and are somehow complementary in their target selection choices.To integrate the regulatory information, we made use of the analogy of this problem with inclusion of the citation information among documents in standard topic modeling applications to texts [23]. In our case, the additional links are not between samples (as it would be a citation link or a hyperlink); therefore, for links between branches in particular, we added gene–miRNA links.We ran the tripartite model as described before; then, in a second moment, we added links gene–miRNA from regulatory network (we tested separately MirDip [61] and TarBase [62]), as shown in Figure 10a. On the fitted triSBM model, we ran steps of the fast merge-split implementation of SBM [63] to improve the description length (see Methods for a precise definition) of the data made by the model, taking advantage of the gene-regulation information in a way similar to the citation between documents when they are used to improve the classification ability of hSBM in that context.We report in Figure 10b the Normalized Mutual Information, measuring the ability of the full process (fit triSBM, add links, run merge-split) in identifying the breast subtypes. Remarkably enough, we see that by including the information on miRNA–genes interactions, we reach a higher NMI, i.e., a better agreement of our clusters with the subtype organization. This does not happen when simply running merge-split after triSBM is run.This shows that it is possible to integrate not only multiple layers of sample-related information, but also knowledge about correlations between different kinds of features. Our results represent a first proof of concept in this direction, and we plan to further pursue this type of analysis in future.As we discussed in the introduction, the nSBM algorithm can be extended in principle to any other layer of information on the samples. A natural candidate is Copy Number Variation (CNV). It is well-known that chromosomal aberrations are a hallmark of cancer and that several types of cancer are characterized by a well-defined set of chromosomal loci whose deletion or duplication can drive the onset of that particular type of cancer. We already noticed that, using the information contained in the miRNA branch, we could identify two loci whose alteration were known to be associated with the onset of breast cancer. In TCGA database, we also have the information on the CNV values for all samples. We included this information by adding a fourth branch to our algorithm (accordingly, we shall call it in the following, “tetraSBM”). As a preliminary test, we selected only genes with positive CNV (i.e., genes contained in duplicated loci) and that were neglected for the moment deletions.We performed a gene selection also in this new branch. Highly Copied Genes were selected, keeping the ones with an average (over samples) CNV greater than 3.5. A total 1353 genes passed our selection. This selection would select genes with at least 2 duplications (CNV=4) on average.It is important to stress that, at this stage nodes, which corresponds with the same gene in the gene expression branch and in the CNV branch, are completely uncorrelated and are seen by the algorithm as independent nodes. We shall discuss below how to address this issue.In our setting, we have 3000 protein-coding genes in the gene expression branch, 1353 genes in the CNV branch, and 417 of them are represented by nodes in both branches.We ran the tetraSBM model on this network with samples, protein-coding genes, miRNAs, and CNV genes and obtained two hierarchical levels. In the first one, the four branches were partitioned into 13 clusters, 7 gene-topics, 5 miRNA-topics, and 5 CNV-topics. In the second one, we found 397 clusters, 49 gene-topics, 14 miRNA-topics, and 31 CNV-topics.Looking at the CNV-topics, we found a very interesting result (see Table 4). Performing the usual Gene Set Enrichment Analysis we found, with very low values of False Discovery Rate (FDR), a few chromosomal loci that we think represent the complete collection of chromosomal aberration associated with breast cancer and could be used as a robust signature of this type of tumor. The relevance of this result is supported by the other set of enriched keywords (taken from [64]), which are reported in Table 4 and show that for some of these loci, the association with breast cancer is already known and is indeed very strong.On the other side, if we test the performance of tetraSBM to identify the samples subtype, we see that, including the information on CNV, we have a decrease in the NMI value (see Supplementary Figure S4). This is not surprising because within the duplicated (or deleted) loci, besides the few drivers of the cancer, there are hundreds of “hitchhikers” genes that simply add noise to the process of subtype classification performed by the other two layers (genes and miRNAs). The variability of the gene expression values that are associated to the different cancer subtypes (and in fact, are allowed to classify the subtypes in the hSBM and triSBM versions of the algorithm) were completely shadowed by the noise induced by the CNV branch. In the Supplementary Figure S5, we reported a bipartite analysis on subtypes with a bipartite network using only the CNV data. This analysis confirms that the CNV layer is less-informative than the layer with only protein-coding genes.This tells us that adding further layers of information does not automatically improve the quality of clustering. It is always important to perform a careful analysis of the biological information contained in the data and of its possible interference with the other layers. In this particular example, we learned that miRNAs cooperate together to assign coregulated genes to the same gene-topic and samples of the same subtype in the same clusters. This fact becomes particularly clear looking at the probability (see Equation (2) in the Methods section and [65] for further details) of moving nodes between groups: when moving a gene between gene-topics, it is more probable to move in a topic where there are genes with many connections to the miRNAs connected to the gene itself. This is confirmed by the fact that, as we discussed in the previous sections, there are miRNA-topics that overlap with clusters of miRNA [56] known to coexpress in breast cancer. On the other hand, the CNV features force samples with the same duplicated loci to be together and this seems not to be correlated with the cancer subtype, at least in TCGA-BRCA data.This does not mean that the addition of CNV data is useless. It is only by including CNV that we may have, as we have seen, precise information on the chromosomal aberrations involved in breast cancer. It is also interesting to notice that this information is somehow complementary to the one we obtained in the previous section looking at the miRNA clusters. The chromosomal loci that we detected there are not present in this CNV analysis because their CNV value is below the threshold we fixed to include CNVs in the tetraSBM.The results published here are, in part, based upon data generated by The Cancer Genome Atlas (TCGA) managed by the NCI and NHGRI. Information about TCGA can be found at https://cancergenome.nih.gov, accessed on 10 February 2022. TCGA data of breast cancer samples were downloaded through portal.gdc.cancer.gov, accessed on 10 February 2022. We selected TCGA program, TCGA-BRCA Project Id, transcriptome profiling as Data Category. We chose Gene Expression Quantification and RNA-Seq as the Data Type and Experimental Strategy to download gene expression data in HTSeq-FPKM. Moreover, we downloaded the number of reads per million of miRNA mapped from the miRNA Expression Quantification Data Type generated with the miRNA-Seq Experimental Strategy.In order to benchmark our results, we compared in Section 2 the clusters of samples obtained by our algorithm with TCGA annotation, which we considered as our “ground truth”. We choose the annotations available through TCGABiolinks [32,33] and, in particular, the one defined as Subtype_Selected. Those subtype annotations are provided by [66] and are based on previously published studies [17,67] about breast cancer based on TCGA.In other analyses, we needed to know if a sample was a primary tumor or derived from normal tissues. Solid Normal Tissues samples are the ones with sample type 10 or 11 in TCGA barcode (10 to 19 are normal types) (https://docs.gdc.cancer.gov/Encyclopedia/pages/TCGA_Barcode/, accessed on 10 February 2022).We downloaded the independent Breast Cancer Consensus Subtypes (BCCS) related to the TCGA files provided by the Supplementary files of [41].We downloaded METABRIC data from the European Genome-Phenome archive.We downloaded METABRIC miRNA landscape study (EGAS00000000122), in particular, Normalized miRNA expression data (EGAD00010000438) and Normalized mRNA expression (EGAD00010000434).We collect here some further information on the nSBM algorithm.
+        The search for optimal allocation of the latent variables is performed by inheriting and expanding [25] hierarchical Stochastic Block Modeling (hSBM) introduced in [10]. Note that the training process is performed simultaneously in all branches of the network: this means that all the types of data contribute to the learning process at the same time, without, in principle, any preference at the beginning.As mentioned in the main text, nSBM attempts to maximize the posterior probability P(θ|A) that the model describes the data
+            (1)P(θ|A)∝P(A|θ)P(θ)
+            in a completely nonparametric [68] way. Instead of maximizing the probability of the model, as usual, it minimizes the Description Length Σ=−logP(A|θ)−logP(θ). We used the minimise_nested_blockmodel_dl function from graph-tool [69]. In our setting, A is a block matrix in which each block is a “Bag of Features” (i.e., genes, miRNAs, …). It can be seen as a two-dimensional matrix whose entries wij are the weights mentioned above. The probability of accepting the move of a node with a neighbor t from group r to group s is [65]
+            (2)Pr→s|t=ets+ϵet+ϵB,
+            where ets is the number of edges between groups t and s; et is the total number of edges connected to group t. From this, another advantage of a multibranch approach should be clear: different ’omics may have their own normalization. In fact, when moving a sample from r to s, the probability is estimated considering only the branch to which t belongs. If the node t is a gene, ets/et is normalized, taking only into account the mRNA expression values.We set the algorithm so as to do a sort of model selection minimizing the Description Length Σ10 times and then choosing the model with the shortest Description Length.We used the nested, degree-corrected [68] version of the model [70] so as to obtain in output a hierarchy of results.The intrinsic complexity of typical Stochastic Block Modeling algorithms is O(nm+τ)E+Vln(V)∗ln(V)ln(σ) (τ, nm, and σ are hyperparameters of the model), which equals O(Vlog2V) if the graph is sparse (E∼O(V)) [71], where V is the number of vertices (samples, genes, and microRNAs) and E is the number of edges. If E>>V, the complexity is not logarithmic and the CPU time needed to minimize the description length increases as well. In this case, to reduce the CPU bottleneck, one can apply a log-transformation to the data, which strongly reduces the number of edges E. We ran the model on a 48-core machine with 768 GB of memory [72].The search for optimal allocation of the latent variables is performed by inheriting and expanding [25] hierarchical Stochastic Block Modeling (hSBM) introduced in [10]. Note that the training process is performed simultaneously in all branches of the network: this means that all the types of data contribute to the learning process at the same time, without, in principle, any preference at the beginning.As mentioned in the main text, nSBM attempts to maximize the posterior probability P(θ|A) that the model describes the data
+            (1)P(θ|A)∝P(A|θ)P(θ)
+            in a completely nonparametric [68] way. Instead of maximizing the probability of the model, as usual, it minimizes the Description Length Σ=−logP(A|θ)−logP(θ). We used the minimise_nested_blockmodel_dl function from graph-tool [69]. In our setting, A is a block matrix in which each block is a “Bag of Features” (i.e., genes, miRNAs, …). It can be seen as a two-dimensional matrix whose entries wij are the weights mentioned above. The probability of accepting the move of a node with a neighbor t from group r to group s is [65]
+            (2)Pr→s|t=ets+ϵet+ϵB,
+            where ets is the number of edges between groups t and s; et is the total number of edges connected to group t. From this, another advantage of a multibranch approach should be clear: different ’omics may have their own normalization. In fact, when moving a sample from r to s, the probability is estimated considering only the branch to which t belongs. If the node t is a gene, ets/et is normalized, taking only into account the mRNA expression values.We set the algorithm so as to do a sort of model selection minimizing the Description Length Σ10 times and then choosing the model with the shortest Description Length.We used the nested, degree-corrected [68] version of the model [70] so as to obtain in output a hierarchy of results.The intrinsic complexity of typical Stochastic Block Modeling algorithms is O(nm+τ)E+Vln(V)∗ln(V)ln(σ) (τ, nm, and σ are hyperparameters of the model), which equals O(Vlog2V) if the graph is sparse (E∼O(V)) [71], where V is the number of vertices (samples, genes, and microRNAs) and E is the number of edges. If E>>V, the complexity is not logarithmic and the CPU time needed to minimize the description length increases as well. In this case, to reduce the CPU bottleneck, one can apply a log-transformation to the data, which strongly reduces the number of edges E. We ran the model on a 48-core machine with 768 GB of memory [72].In our setting, we have V∼O(1000) vertices, E∼O(1000000) edges, and the network is indeed very dense. In order to reduce the number of nodes and edges, a preprocessing step is needed. We shall discuss this issue in the next subsection.We considered 1222 samples from TCGA-BRCA project and selected the 1200 with a valid annotation from [32,33]; then, we ran the model on a tripartite network built with normal and tumor samples from TCGA on one branch, 3000 FPKM normalized gene expression data on a second branch and 1300 miRNA-Sequencing data on the third branch. Note that we did not explicitly selected the known breast Cancer markers, our approach to topic model, as already discussed in [3], took into account the whole expression pattern and did not relay only on few specific markers.The output of the tripartite model consisted of two hierarchical levels with 1 and 14 clusters; 11 and 331 topics; and 33 and 47 miRNA-topics on the three branches, respectively. We ran also, as a comparison without miRNAs, hSBM on a bipartite network and obtained levels with 2, 11, 76, 608 clusters and 5, 17, 62, 390 topics across the hierarchy.As the output of the model, we find the probability distributions P(topic|sample) and P(gene|topic). These probabilities are defined, in terms of entries of the program, as follows:(3)P(topic|sample)=numberofhalf-edgesonsamplecomingfromtopicnumberofhalf-edgesonsample
+        and
+        (4)P(gene|gene-topic)=numberofhalf-edgestogene-topicgoingtogenenumberofhalf-edgestogene-topic.The same is true for miRNA-topics and for each and every eventual additional layer of features.The data provided in the atlas consisted of 1222 (∼1100 have both mRNA and miRNA transcript profiles data) samples associated with almost 20,000 genes and 2000 miRNA entries. Without preprocessing, this would have led to an adjacency matrix too big to be handled efficiently by the algorithms.We performed two kinds of preprocessing to reduce the number of nodes and the number of edges.In order to reduce the number of nodes, we filtered genes and miRNAs selecting only the highly variable ones. The highly variable are the ones with the highest dispersion (variance over mean) with respect to the genes with the same average expression. This selection was performed using the scanpy python package [73]. This analysis was performed separately on genes and microRNAs since they are provided by different experiments and different normalization. We selected in this way 3000 genes and ∼1200 miRNAs.Furthermore, we applied a standard approach to reduce the weights of the links and applied a log(FPKM+1) transformation to the data before running the topic models. This helped us to reduce by some order of magnitudes the number of edges (as we mentioned above, in this class of algorithms, the weight of a link is mimed adding multiple edges with weight 1) and the model ran several times faster.In the Copy Number Variation analyses, we chose ∼1300 genes with an average CNV >3.5.An interesting feature of the SBM type of algorithm is that they are typically robust with respect to gene selection. In the analyses of this paper, we considered only highly variable genes; however, in the supplementary material of [3], we discussed different types of gene selections showing that they were typically leading to similar performances.In the analysis of the METABRIC dataset, we utilized the previously selected genes and microRNA.To evaluate the agreement between the sample partitions and the annotations, we chose the so-called “Normalized Mutual Information” (NMI), which was proposed in [40], in a new evaluation framework for topic models. Moreover, as discussed in [3], it can be shown that NMI is the harmonic average of two metrics that evaluate, respectively, the completeness and the homogeneity of a partition of annotated samples [39]. A cluster is complete if all samples with a given label are assigned to the same cluster; a partition is homogeneous if, in a cluster, all the samples have the same annotation. In order to correctly identify the cancer subtype of a given sample, one would prefer to have a partition in clusters that is both homogeneous and complete.The NMI is estimated using Shannon’s entropy formula to measure the quantity of information in the partition. The problem of this measure is that even in a random partition, there is a residual entropy and the NMI is not zero; this effect is particularly important in the layers of the models with high resolution (many clusters). In order to avoid this bias, we evaluated this default NMI by randomizing the subtype annotations of the samples. This was performed multiple (∼50) times, each time preserving the number of clusters and the number of samples in every cluster; we call the average NMI on these multiple random assignments NMI∗; this is the residual information on the considered partition. In the results, we reported NMI/NMI∗, which measures how much information the model learns with respect to a random assignment. It is important to stress that this measure has no absolute value and should not be used to compare performances on different datasets; however, it can be successfully used to compare different algorithm in the same dataset, which is what we did in the Results section.In addition to the NMI, it is also possible to compare different classes of topic modeling algorithms on their ability to compress the data [65,74]. This can be addressed measuring the description length Σ of a model, which represents, in nat units, the number of bits a model requires to describe the data network. Unlike NMI, it has the advantage not to rely on any ground truth. Using ΣE (where E is the total number of edges), it is possible to measure the quantity of information that the model requires to describe an edge. In the models of Figure 5, hSBM requires a ΣE∼6,26, which is greater than the 1,4 units required by triSBM. One can estimate the difference of the two ΔΣE≃4,9; this can be related to the Bayes factor [75] (being the posterior P=exp−Σ) Λ=expΔΣ≃e4,9≃102,1, meaning that the model with miRNA is a ∼100 times more probable description of the data network links. The description lengths of the tetrapartite model and the model with regulatory network are reported in Supplementary Materials (see Figure S6).From the P(topic|sample) distributions, it is easy to obtain the probability P(topic|subtype) by averaging over all samples belonging to the same subtype.Then, by subtracting to P(topic|subtype) the mean value over the whole dataset, we find a new set of quantities that we define as “centered” distributions (we already used them in [3]; they have the same meaning of the normalized value of the mixture proportion τ in [23])
+        (5)P¯(topic|sample)=P(topic|sample)−1RΣs∈samplesP(topic|s),
+        where R is the total number of samples. This procedure can be implemented separately both on the miRNA-topic and on the gene-topic side. The centered P(topic|sample) can be represented as box plots, after grouping samples by their subtype. Examples of these are the box plots reported in Figure 6 on the gene side and Figure 7 on the miRNA side.We performed the survival analyses fitting a COX [58] model.Our analysis began with the list of the mixtures P(topic|sample). We cleaned up the stages’ labels, removing any additional letter (e.g., stage ia became stage i), and ended up with four stages: i, ii, iii, and iv.Using Genomic Data Commons tools, we downloaded TCGA metadata and, in particular, demographic.vital_status, demographic.days_to_last_follow_up, demographic.days_to_death, demographic.gender, and diagnoses.age_at_diagnosis. We estimated the lifetime or the number of days the patient survived after the diagnosis, using days_to_last_follow_up if the patient was Alive and days_to_death for Dead patients. A similar approach was recently utilized by [76].In order to estimate whether a topic is upregulated in a patient, we evaluated the 35th percentile of P(sample|topic) and considered it as a threshold thr. Then, we engineered a feature as follows:(6)up(sample)=1P(topic|sample)>thr0P(topic|sample)≤thrWe used these data to fit the hazard with a COX model. These analyses were performed using the lifelines Python package [77] and, in particular, the COXPHFitter module. We used the lifetime, vital status, and the new feature as input for the fit function.The Cox model quantified how the topic of miRNAs regulation affected the survival probability. Cox fits the hazard function conditioned to a variable h(t|x)=b0(t)∗eΣi=0nbi∗(xi−x¯i). x is the vector of the n covariates considered. The hazard is defined as the ratio of the derivative of the survival and the survival itself h(t)=−S′(t)S(t). S(t) is the probability of being alive at time t, namely, the number of patients alive at time t divided by the total number of patients. The package estimated the ratio between the hazard of samples with topic upregulated and hazard of samples with topic not upregulated. Therefore, we were able to estimate the exp(coef) or hazard ratio exp(coef)=hazardofsampleswithtopicup-regulatedhazardofsampleswithtopicnotup-regulated. Note that the coef does not depend on time but is a sort of weighted average of period-specific hazard ratios.The Python package to run nSBM [25] can be downloaded from GitHub ( https://github.com/BioPhys-Turin/nsbm, accessed on 10 February 2022) or, alternatively, can be installed using Anaconda (https://anaconda.org/conda-forge/nsbm, accessed on on 10 February 2022) by running conda install nsbm -c conda-forge.We discussed in this paper the application using genomics data; however, the package is written in a way that makes it agnostic with respect to the type of data it receives in input and to the number of branches. One can ideally integrate as many different sources (’omics) of data as needed. Eventually, it can process not only biological data, but every kind of dataset whose input could be represented as a rectangular matrix (Bag of Words) for each feature.In conclusion, the nSBM model we propose here, integrating multiple sources of information into an hSBM analysis, should be useful to extract a lot of information from transcriptomics data.
+      Using the python package: nSBM, inherited from hSBM [10], ready to install and easily executable on n-partite networks, will be straightforward to address different types of biological data.Second, the integration of multiple sources of data, such as microRNA expression levels and the protein-coding mRNA ones, greatly improves the ability of the algorithm to identify breast cancer subtypes.Third, we use our results to identify a few genes and miRNAs and characterize a few chromosomal duplications that seem to have a particular prognostic role in breast cancer and could be used as signatures to predict the particular breast cancer subtypes.Using the python package: nSBM, inherited from hSBM [10], ready to install and easily executable on n-partite networks, will be straightforward to address different types of biological data.Second, the integration of multiple sources of data, such as microRNA expression levels and the protein-coding mRNA ones, greatly improves the ability of the algorithm to identify breast cancer subtypes.Third, we use our results to identify a few genes and miRNAs and characterize a few chromosomal duplications that seem to have a particular prognostic role in breast cancer and could be used as signatures to predict the particular breast cancer subtypes.In conclusion, this paper released a new tool to easily integrate different sources of data into a topic-modeling analysis.We showed some application in a specific case (breast cancer) with some sources of data (mRNA, miRNA, CNV). Indeed, this approach can be applied to other datasets and, more importantly, to any possible sources of data (genomics, proteomics, lncRNA, circRNA…).The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051150/s1, Figure S1: Normalized Mutual information of hSBM and triSBM partition compared with the Breast Cancer Consensus Subtypes of Ref. [41]. Figure S2: Validation on METABRIC dataset. Figure S3: Multivariate (Log) Hazard Ratios. Figure S4: Normalized Mutual Information of models with samples and mRNA (hSBM), miRNA (triSBM) and mRNA, and both miRNA and CNV (tetraSBM). Adding CNV introduces noise to the model. Figure S5: Normalized Mutual Information of bipartite models with samples and mRNA (hSBM) and samples with Copy Number Variation (CNV). Adding CNV introduces noise to the model. Figure S6: Description length of different settings. Figure S7: Days of survival of different patients in clusters.Conceptualization, F.V., M.O. and M.C.; methodology, F.V., M.O. and M.C.; software, F.V.; writing—original draft preparation, F.V. and M.C.; writing—review and editing, F.V., M.O. and M.C.; visualization, F.V. All authors have read and agreed to the published version of the manuscript.This work was partially supported by the “Departments of Excellence 2018–2022” Grant awarded by the Italian Ministry of Education, University and Research (MIUR) (L.232/2016).Notebooks to reproduce the results in this work are available on GitHub at https://github.com/BioPhys-Turin/keywordTCGA, accessed on 10 February 2022).We would like to acknowledge the Competence Centre for Scientific Computing C3S which provided us the access to the computing cluster OCCAM. The results shown here are, in part, based upon data generated by TCGA Research Network: https://www.cancer.gov/tcga, accessed on 10 February 2022.The authors declare no conflict of interest.The following abbreviations are used in this manuscript:
+      SBMStochastic Block ModelingTCGAThe Cancer Genome AtlasGSEAGene Set Enrichment AnalysisFDRFalse Discovery RateFPKMFragments Per Kilobase of transcript per Million mapped readsCartoon of multipartite networks with samples, protein-coding genes, and microRNAs. (a) A bipartite network with a layer of protein-coding genes and a layer of samples. A gene is connected to a sample if it is expressed in that sample and the link weight is proportional to the expression level. (b) A tripartite network obtained by adding the miRNA expression layer. The topic model algorithm essentially outputs a block or topic structure in each layer.Adding miRNA leads to new topics. The Adjusted Mutual Information between the outputs of triSBM and hSBM (i.e., with and without miRNA). The partitions obtained in output are different for any combination of layers.Clustering of breast samples with and without the miRNA branch. We compare normal and solid tumor tissues from TCGA using (a) triSBM and (b) hSBM at a similar resolution level.The increase in performance when separating tumor and normal samples by the addition of the miRNA layer. The NMI is evaluated at different resolution levels (numbers of clusters) using (triSBM) or not using (hSBM) the information of miRNA expression. The normal/tumor annotation from TCGA is used as ground truth.Scores and partitions based on Subtype_Selected annotations from [32,33]. (a) Scores for both (triSBM and hSBM) setting for the subtype classification problem. (b) The miRNA are introduced. We compared the two settings choosing the layers with a compatible number of clusters. (c) The clusters from a simple bipartite setting. They are almost similar; in (c), Luminal B is identified better. We define Normal as the Solid Tissue Normal from TCGA, whilst Normal-Like are the Primary Tumors annotated BRCA.Normal from [32].Box plots of the centered P(gene-topic|sample) for different gene-topics. Samples belonging to each subtype may be over- or under-expressed in different topics.Box plots of the centered P(miRNA-topic|sample). This plot shows that the differences of topic expression in each subtype may be different. Some miRNA-topics are more abundant in samples known to be Basal Subtype.Genes that are annotated in the Gene Set Enrichment Analysis terms contribute more than average to the topic. Contribution of miRNAs to miRNA-topic 7. miRNAs that belong to the ontology specific of locus c14q32 are highlighted and have high P(miRNA|miRNAs’topic).Kaplan–Meier analysis of miRNA-topic 7. We divided patient (samples) into two cohorts using the information regarding the importance of this miRNA-topic in each sample. Patients with a great presence of these topics have smaller values of survival.Configuration and scores when adding gene–miRNA links. (a) A graphic of a tripartite network with links gene–miRNA. (b) The scores of this new setting using two different (mirDIP [61] and TarBase [62]) regulatory networks separately.GSEA FDR enrichment P-values on the gene-topics. For each gene-topic, only the terms with the strongest enrichment are reported. In brackets is the number of genes in each set (topic). Lists are available at https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/trisbm/trisbm_level_0_topics.csv, accessed on 10 February 2022.GSEA results on the miRNA-topics. We selected and reported the ones with the strongest enrichment. Lists are available at https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/trisbm/trisbm_level_1_metadata.csv, accessed on 10 February 2022.microRNAs sorted by their P(miRNA|miRNA-topic7). The most important miRNAs in our candidate miRNA-topic. Most of them are well-known in literature. The complete list is available at https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/trisbm/trisbm_level_1_keyword-dist.csv, accessed on 10 February 2022.Enrichment analysis on the Copy Number Variation branch of tetraSBM. All the lists are available at https://github.com/BioPhys-Turin/keywordTCGA/blob/main/brca/tetrasbm/trisbm/trisbm_level_0_kind_3_metadata.csv, accessed on 10 February 2022.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01151.txt

@@ -0,0 +1 @@
+The development of new strategies for the management of cutaneous metastases is a major clinical challenge because of the poor prognosis. To advance in this field, a better understanding of the molecular alterations involved in the metastatic process is needed. In the present study, the clinicopathological characteristics of breast cancer that develop cutaneous metastases were analyzed and the molecular differences between primary breast tumors and their corresponding cutaneous metastases were compared. We observed that the surrogate molecular type of breast cancer with an increased risk to metastasize to the skin was triple negative. In total, 48.5% of the cutaneous metastases presented some additional molecular alteration with respect to the primary tumor. However, no characteristic mutational pattern related to skin metastasis development was observed. Identifying the genes involved in the development of cutaneous metastases is important to gain insights into the biology of the disease and to identify possible diagnostic and therapeutic biomarkers.Background: The characterization of molecular alterations of primary breast carcinomas (BC) and their cutaneous metastases (CM) to identify genes involved in the metastatic process have not yet been completely accomplished. Methods: To investigate the molecular alterations of BC and their CM, a total of 66 samples (33 BC and 33 CM) from 33 patients were analyzed by immunohistochemical and massive parallel sequencing analyses. In addition, the clinicopathological characteristics of patients and tumors were analyzed. Results: Triple negative (TN) BCs were overrepresented (36.4%) among tumors that developed CM. A change of tumor surrogate molecular type in metastases was found in 15% of patients and 48.5% of the CM presented some additional molecular alteration with respect to the primary tumor, the most frequent were amplification of MYC and MDM4, and mutations in TP53 and PIK3CA. Survival was related to histological grade, tumor surrogate molecular type and TP53 mutations in the univariate analysis but only the tumor surrogate molecular type remained as a prognostic factor in the multivariate analysis. Conclusions: The TN molecular type has a greater risk of developing skin metastases. There are phenotypic changes and additional molecular alterations in skin metastases compared to the corresponding primary breast tumors in nearly half of the patients. Although these changes do not follow a specific pattern and varied from patient to patient, they could impact on the treatment. More studies with larger patient and sample cohorts are needed.Breast cancer (BC) is the most prevalent malignancy in females and is the leading cause of cancer death in women [1]. We can distinguish between different groups of BC according to the molecular profile as those that express Estrogen Receptor (ER) and/or Progesterone Receptor (PR) (75%), those that express Epidermal Growth Factor Receptor 2 (Her2) (15%) [2], and tumors that do not express any of these three markers, the triple-negative (TN) tumors (10–20%) [3]. According to these characteristics, we can apply a surrogate molecular classification that distinguishes four surrogate molecular types of BC: luminal A-like, luminal B-like (HER2− and HER2+), HER2+ (non-luminal), and triple negative (TN) [4].Metastasis accounts for the majority of deaths from BC [5]. It is a complex process in which the cells of a primary tumor are propagated to distal organs, showing uncontrolled growth in these tissues [6,7]. Normally, BC metastasize to the lung, bone, and brain [8]. Moreover, BC is the tumor most prone to develop skin metastases in females [9].Skin metastases are the result of lymphatic embolization, hematogenous or contiguous spread [10] and are present in around 24% of patients with metastatic BC [9,11,12,13,14,15,16]. In addition, due to the high incidence of BC, these skin manifestations are the most common metastases among women seen by dermatologists, specifically, 69% of these metastases come from BC [13].Cutaneous breast metastases tend to develop in the vicinity of the primary tumor in the skin of the breast and chest wall, although they can also develop in the abdomen, extremities, head, or neck [9,12,13]. In addition to being able to develop in different locations, breast cutaneous metastases (CM) can manifest in a wide variety of ways. Nodules are the most common presentation (80%), but there are other patterns [11,13] such as telangiectatic pattern with an incidence of 8–11%, erectile pattern with an incidence of 3–6.3%, carcinoma en cuirasse with an incidence of 3–4% and neoplastic alopecia with an incidence of 2–12% [11,13].All BC metastases, regardless of subtype, usually occur late in the disease, in the later stages of the disease course. Brownstein et al. [17] observed that skin metastasis was the presenting sign of the disease in only 3% of cases of metastatic BC. Kong et al. [15] observed that 56.8% of the patients had more than one visceral metastasis at the time of diagnosis of CM. Therefore, when diagnosed, the primary tumor is widespread and may not be treatable (palliative care, surgical excision, or complete mastectomy is provided) [18]. All this confers a poor prognosis, with an average survival of 3 to 6 months, with few differences regarding whether the lesions are single or multiple, with mortality exceeding 70% in the first year after diagnosis [19].Since there are few series evaluating the molecular alterations of CM and most do not include the analysis of both the primary tumor and CM, the objective of this study was to compare the molecular alterations of matched primary BC and their CM in a series of 33 patients to better understand the genes implicated in BC progression and to identify potential therapeutic targets.Compliance with Ethical Standards: All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and national research committee and the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The use of patient samples to meet the objectives of this study was approved on 14 May 2021 by the Ethics Committee of the Hospital Ramón y Cajal (ethical approval code: 30-21).Histological sections of all primary tumors and their metastases were reviewed by two experienced pathologists (J.P and B.P.-M.). Histological typing and grading was performed according to WHO recommendations [4]. Lymphovascular invasion (LIV) was also evaluated in primary tumors. Cutaneous lesions were diagnosed as metastases from BC based on a biopsy, excluding cases with direct extension from a subjacent breast lesion.All 33 primary tumors and 33 metastases underwent an immunohistochemical study for the expression of estrogen receptors (ER), progesterone receptors (PR), HER2 and Ki67. Immunostaining was performed using the EnVision detection system (K5007, Agilent Dako, Glostrup, Denmark) using the following antibodies: ER (clone EP1, Agilent Dako Omnis, Glostrup, Denmark), PR (clone PR 1294, Agilent Dako Omnis, Glostrup, Denmark), HER2 (SK001, clone poly, Agilent Dako autostainer, Glostrup, Denmark), and Ki-67 (clone MIB-1, Agilent Dako Omnis, Glostrup, Denmark). Evaluation of ER, PR, and HER2 expression was performed according to American Society of Clinical Oncology and the College of American Pathologists (ASCO-CAP) guidelines) [20]. HER2 equivocal cases (2+) underwent FISH analysis, using the PathVysion HER-2 DNA Probe Kit (PathVysion II kit, Abbot Laboratories, Abbot Park, IL) on complete tumor sections. Results were interpreted according to 2018 ASCO-CAP guidelines [20].In the 7 invasive lobular carcinomas, the expression of E-cadherin (clone NHC-38, Agilent Dako Omnis, Glostrup, Denmark) was studied to confirm the histological type. Androgen receptor (AR) (SP107 Cell Marque, Ventana Medical Systems, Rocklin, CA, USA) was also determined to evaluate the possible apocrine phenotype in 12 primary TN tumors and their metastases.Tumors were classified into different surrogate molecular types as Luminal HER2-, Luminal HER2+, HER2 (non-luminal), and TN.Sufficient DNA for sequencing was obtained from the 33 primary breast carcinomas and their corresponding CMs. Areas with >30% of tumor cells were obtained by “punching” paraffin blocks in selected areas previously marked on hematoxylin/eosin (H&E) slides. The QIAamp DNA FFPE Tissue Kit (Qiagen, Valencia, CA, USA) was used to extract DNA from all samples. Quality of DNA samples was measured using TapeStation (Agilent 2200 TapeStation, Santa Clara, CA, USA), whereas quantification was performed by QUBIT 2.0. (Thermo Fisher Scientific Qubit 2.0 Fluorometer, Waltham, MA, USA).A custom gene panel was designed using the SureDesign platform by Agilent Tech. (Santa Clara, CA, USA) to consistently target 61 genes (AKT1, ARID1A, ARID1B, ARID5B, ATR, BCOR, BRCA1, BRCA2, BRAF, CASP8, CCNE1, CDH1, CDH4, CDH19, COL1A1, CSMD3, CTCF, CTNNB1, EGFR, ERBB2, ESR1, FBXW7, FGFR1, FGFR2, FOXA2, GRB7, GSDMB, MAP2K4, KRAS, MAP3K1, MLL3, MLH1, MKI67, MSH2, MSH6, MYC, NCOR1, NF1, NRAS, PGAP3, PIK3CA3, PIK3R1, PMS2, PNMT, POLE, PPP2R1A, PRPF18, PTEN, KMT2B, RB1, RPL22, SF3B1, SPOP, STARD3, TAF1, TBX3, TCAP, TP53, VGLL1, ZNF217, ZNF703). For library construction, a modified protocol for Agilent SureSelectXT FFPE was selected based on Covaris AFA fragmentation of DNA (Covaris, Woburn, MA, USA) and subsequent probe-mediated hybridization capture. Sequencing of equimolar libraries was performed using the Ion S5™ Torrent (Thermo Fisher Scientific, MA, USA).Bioinformatics analysis was carried out using a specific pipeline using Novoalign V3 (2021) (http://www.novocraft.com/products/novoalign/ accessed on 4 February 2022) as aligner and VarScan [21] as variant-caller, with no filters. Variant annotation was performed using the VEP from Ensembl version 88 (http://www.ensembl.org/info/docs/tools/vep/index.html accessed on 4 February 2022), which corresponds to the hg38 version of the human reference genome. Variants were latterly filtered using the functional information (taking only deleterious variants), the variant allele frequency (>0.05), and the strand-bias from both the variant and the reference allele. If normal tissue was available, those variants also present in the normal component were excluded. Finally, visual inspection was performed as the final selection criterion using the IGV browser [22].In addition, 11 mutations were confirmed by Sanger sequencing, 3 in the PIK3CA gene (Pt3, Pt14 and pt36), 6 in the TP53 gene (Pt1, Pt2, Pt4, Pt10 and Pt11), and 2 in the ERBB2 gene in the samples (pt13). (Supplementary Table S1).Since our panel was not designed to detect CNVs, a tissue microarray (TMA) was constructed to evaluate gene copy number variations in CCND1, MYC, FGFR1 and MDM4, the genes most frequently amplified in BC, by Fluorescent In-Situ Hybridization (FISH). Only 20 matched primary tumors and metastases (40 samples) were included in the TMA due to sample limitation after the initial immunohistochemical/molecular study. Chromosomal alterations were evaluated by FISH on TMA sections using the following probes: SPEC CCND1/CEN11, MYC/CEN8, FGFR1/CEN8, and MDM4/1p12 dual color Probe Kit (Zytovision GmbH, Bremen, DE). FISH slides were observed with a fluorescence microscope at 100X with immersion oil. A detailed scoring of at least 20 neoplastic cells per sample was performed. Amplification was considered when the tumor cell population had at least twice as many gene signals than centromere signals of the respective chromosome (ratio ≥ 2), and polysomy when the average of centromere signals on tumor cells were >3.The Kaplan–Meier method was used to calculate overall survival according to clinicopathological characteristics (age, pT, pN, histological type, surrogated molecular type, LIV, histological grade, clinical stage, metastasis location, and neoadjuvant therapy) and mutations in TP53 and PIK3CA genes. Cox proportional hazards models were used to investigate the association between mortality and clinicopathological and molecular features.A total of 33 patients diagnosed with BC and CM between 2005 and 2020 from the Pathology Department in Ramón y Cajal University Hospital (Madrid, Spain) and in 12 de Octubre University Hospital (Madrid, Spain) were selected, all had available paired samples (primary tumor and CM).Clinicopathological features of all 33 primary samples are presented in supplementary Table S2 and summarized in Table 1. The median age of the patients at diagnosis was 63.5 years old (range 29 to 84), 57.6% of patients were older than 60 years.According to the immunohistochemical profile, 16 cases (48.5%) were Luminal HER2- surrogate molecular type, 3 cases (9%) were Luminal HER2+, 2 cases (6%) were HER2+ (non-Luminal), and 12 cases (36.4%) were TN.Examining the distribution of histological types, 24 cases (72.7%) corresponded to invasive breast carcinomas of non-special type (BCNST), the rest were special histological types. The largest group of 7 cases (21.2%) corresponded to invasive lobular carcinoma (ILC), one case (3%) was a matrix-producing (chondroid) metaplastic carcinoma, and one case was an apocrine carcinoma (3%).The surrogate molecular type of the tumors and metastases was confirmed by IHC and FISH. The most frequent type of primary tumor was Luminal HER2- followed by TN. The molecular type changed between the primary tumor and its CM in 5 patients (15%), the most common being from luminal to TN (Table 2).In case Pt14, HER2 amplification was lost in the metastasis; on the other hand, in case Pt32, HER2 was overexpressed in the metastasis due to polysomy of chromosome 17.AR expression was studied in 12 TN primary tumors and their metastases by IHC, since AR expression in TNBC is related with the apocrine molecular type [23]. In this series, 3 out 12 TNBC expressed AR and expression was concordant in primary tumors and the associated metastases.The following molecular analysis is based on 66 samples (33 primary tumor and 33 CM) for mutation analysis and 40 samples (20 primary tumor and 20 CM) for CNV analysis. Figure 1 shows the mutations and CNVs found in 29 pairs. In 4 cases, no mutations or CNVs were found, and all were Luminal HER2- cases. Molecular alterations were detected in 12 TN, 12 Luminal HER2-, 3 Luminal HER2+ and 2 HER2+ (non-Luminal) primary tumors. The number of mutations ranged between 1–5. A summary table with the type of mutation and CNVs found in each of the samples is presented in supplementary Table S3.Among the 33 matched cases, TP53 was mutated in 13 primary tumors (39.4%) and in 14 CMs (42.4%). PIK3CA was mutated in 13 primary tumors (39.4%) and 15 CMs (45.4%). NF1 was mutated in 3 of the primary tumors (9%) and 4 CMs (12.1%). AKT1 was mutated in 3 primary tumors (9%) and 3 CMs (9%). ERBB2 was amplified or there was polysomy in 5 primary tumors (15.15%) and 5 CM (15.15%). Among 20 matched cases, MYC was amplified in 3 primary tumors (15%) and 5 CM (25%). MDM4 was amplified in only 3 CM (15%). FGFR1 CNVs (amplification or polysomy) were observed in 2 primary tumors (10%) and 2 CMs (10%). Finally, CCND1 was amplified in 1 primary tumor (5%) and 1 CM (5%). Table 3 shows a summary of these alterations distributed by surrogate molecular types.Analyzing TP53 and PIK3CA mutation frequencies in the CMs of BC cases diagnosed at early stages (I-II) versus those diagnosed at advanced stages (III-IV), we observed that they were very similar. The mutation frequency for both genes of cases diagnosed at early stages was 44.4%. In cases diagnosed at advanced stages the frequencies were 47.8% for TP53 and 52.2% for PIK3CA. Thus, there were no statistically significant differences between the two groups.Additional molecular alterations were observed in the CM, either mutations or CNVs, in 16 patients (48.5%) (Figure 2). In 7 cases (21.2%) there was more than one additional alteration in the CM. Table 4 shows the distribution of these alterations by surrogate molecular types.There were some differences in the frequency of additional mutations depending on the local or distant nature of the CM. In the 17 paired cases that developed distant CM, additional alterations in the CM with respect to the primary tumors were found in 11 cases (64.7%) (5 TN, 4 Luminal HER2-, and 2 HER2+). In contrast, among the 15 paired cases that developed local CM, only 5 cases (33.3%) (3 TN, 1 Luminal HER-, and 1 HER2+) showed additional molecular alterations. (Table 4).In addition, 11 mutations were confirmed in the TP53, PIK3CA, and ERBB2 genes. Figure 3 shows the different mutations found in the ERBB2 gene (L755S and S310F) between the primary tumor and the MC of Pt13, as well as the verification by Sanger.In our series, 7 patients were alive and 26 had died when the data were censored. The median survival since the diagnosis of the disease was 53 months and the median survival since the diagnosis of the CM was 19.6 months. Of the patients who developed distant CM, the median survival was 14.5 months, with an overall survival in the first year of 46.7% of patients. In contrast, when MC was local, the median survival was 34.6 months, with an overall survival in the first year of 40% of patients.The association between overall survival since the diagnosis of the primary tumor and age, location (local or distant), pT, pN, clinical stage, surrogate molecular type and histological type, histological grade, LIV, and neoadjuvant therapy were assessed. In addition, the association between overall survival and TP53 and PIK3CA mutations was assessed. Histological grade, surrogate molecular type, and TP53 mutations significantly affect overall survival (p = 0.015, p = 0.0011 and p = 0.019, respectively) (Figure 4a–c). The shortest overall survival was observed in the TN surrogate molecular type (Figure 4a). By multivariate analysis with the 3 significant variables, the only independent variable in the CoxPh analysis was surrogate molecular type, where HER2- Luminal was the best prognostic type. (Figure 4d).In contrast, the variables pT, pN, stage, permeation, age, histological type location (local or distant), neoadjuvant therapy, or PIK3CA mutations did not significantly affect overall survival.In this study, the clinicopathological features were analyzed of 33 females with BC that developed cutaneous metastases, distantly in 53.1% and locally in 46.9% of the patients. Our observation that 46.9% of CM developed locally on the skin of the breast/thorax is in accordance with the review by Johnson et al. [24], who found in a study of 61 patients with CM, 57% showed metastases in the breast/thorax skin.In the present series, CM developed at a median of 22.8 months after the initial diagnosis of the primary BC. CM usually appear at the end of the disease, late during cancer evolution. Thus, Lookingbill et al. [25] found that only 6.3% of patients with BC had cutaneous involvement at the time of diagnosis of the primary tumor. More recently, Johnson et al. [24] reported that approximately 13.7% of the patients in the 9 retrospective series reviewed [26,27,28,29,30,31,32,33,34] had a skin lesion before or simultaneously with the diagnosis of BC. Supporting these observations, we found that 18.2% of patient in the present series debuted with skin involvement.The median age of women with BC who developed CM varies among series. Whereas the age in our series (63.5 years) was similar to the Johnson et al. [24] review of 41 patients, other studies have reported a mean of 74 years (n = 18) [9] and 48 years (n = 125) [15]. These differences may be related with the different number of patients included in each series or due to ethic/geographic differences, since these series originated from Italy, South Korea, and Spain.In the present series, the histology of 24 cases (72.7%) corresponded to BCNST, and the rest to special histologic subtypes. Among them, the largest group corresponded to ILC (21.2%). However, considering only the 16 Luminal HER2- cases in our series, 43.7% were ILC. Since the frequency of ILC in Luminal HER2- BC is between 15% and 20% [35], our results suggested that ILC is a major risk factor for developing skin metastases. In accordance with our results, the study of Li et al. [36] that included only metastases from Luminal HER2- BC, reported that the proportion of ILC was 26% in the group of patients that developed CM.We observed that the frequency of BC developing CM varied according to the subtype. Table 5 compares the data from our series with those from other published series [15,37,38,39], including our recently published CM review that includes samples (n = 58) of molecularly characterized CM [39,40,41,42,43].Although there were differences between series, a finding common to all of them is the relative overrepresentation of TNBC, since its frequency in the general population of BC is around 15% [44], but it was between 23–39% in BC with CM, suggesting that this surrogate molecular type could be associated with a greater potential to metastasize to the skin.In our series, 15.2% of BC that developed CM were HER2+, a frequency similar to the general population of BC, at least in Spain, suggesting that this surrogate molecular type does not have a special propensity to develop CM. However, Table 5 shows important differences in the frequency of HER2+ BC that develop CM among different series. These differences are probably due to differences in sample size, patient selection, methods of HER2 detection and the period of study.In this series, we studied the mutational landscape of 33 matched primary tumors with their corresponding CM by NGS. In addition, changes in copy number of CCND1, FGFR1, MDM4 and MYC were analyzed by FISH in 20 paired samples. The molecular landscape of primary tumors in this series was concordant with many previous reports demonstrating different mutational patterns among different surrogate molecular types. Thus, whereas mutations in PIK3CA predominated among Luminal BC, TP53 mutation was the main molecular alteration in TNBC.When primary tumors were compared with their respective metastasis, we found that 48.5% of CM exhibited additional pathogenic mutations and/or gene amplification in important oncogenes and/or tumor suppressor genes. Genes involved in the progression of more than one case in the present series included MYC, MDM4, PIK3CA and ERBB2. In spite of this high frequency of additional changes, we did not observe a specific mutational pattern related to tumor progression, indicating that CM is a very individual process in each tumor. These results add to the observations of three previous series analyzing paired primary BCs and their CMs including a total of only 15 patients [39] (Table 6).There are several studies analyzing molecular alterations in metastatic BC including samples from CM, but without comparison with their primary tumors (see González- Martínez et al. [37]). In general, these studies also confirmed that there is no specific pattern of mutations that predispose to CM. Only Rinaldi et al. [39] observed that alterations in NOTCH1 were overrepresented in CM when compared to other metastatic locations.We observed that the number of CM with additional molecular alterations was higher in distant than local metastases. Thus, additional mutations and CNVs were observed in 11 out of 17 (64.7%) distant CM but in only 5 out 15 (33.3%) local CM. These differences seemed to indicate a more advanced molecular stage of distant CM, although they did not have prognostic implications.It is also worth noting that there was hardly any difference between the mutation frequency of TP53 and PIK3CA in CM samples of patients diagnosed at early and advanced stages, so we assume that treatment has no effect on the mutation pattern.An important question in the study of cancer metastasis is whether or not the molecular characteristics of metastatic samples modify patient treatment. In 5 patients in this study (15.1%), there was a conversion of the surrogate molecular type between the primary tumor and its CM. In 3 cases (9%), there was a change from luminal to TN, from Luminal HER2+ to Luminal HER2- in one case (3%), and from Luminal HER2- to Luminal HER2+ in another case (3%). All these changes would have impacted on the treatment decision regarding the use of hormone therapy or anti-HER2 therapy.Regarding specific mutations, three CM developed additional PIK3CA mutations, which is a target for treatment with alpelisib in RH+ metastatic BC. However, the three primary and metastases were TN. We detected an ESR1 mutation in a primary HR+ tumor and the corresponding CM. ESR1 mutations are the most frequent additional mutations that develop in metastatic HR+ BC after hormone therapy, especially after the use of aromatase inhibitors, being infrequent in primary tumors. In this case, ESR1 mutations would have influenced the response to hormone therapy during the complete evolution of the disease.An interesting case in this series was the tumor and the CM developed in patient Pt13, which was an ILC that carried the pathogenic L755S mutation in the primary tumor. This mutation was absent in the CM, which carried the pathogenic mutation S310F. ERBB2 mutations are more frequent in ILC (6%) than in IDC (1.5%), especially in ILC with pleomorphic features, and are associated with a poor prognosis in ILC [45,46]. In addition, response to different anti-HER2 therapies differed among mutations. Thus, whereas L755S seems to be resistant to trastuzumab and lapatinib but sensitive to neratinib and afatinib, S310F seems to be sensitive to all drugs.Patients with CM have a very poor prognosis. Lookingbill et al. [47] observed an average survival of 31 months after the diagnosis of the CM. Kong et al. [15] observed a median survival of 32 months. In our series, the median survival was 19.6 months, and this difference may be related to the clinical differences between the series. Additionally, patients who developed distant CM had a shorter overall survival and died more frequently during the first year after the diagnosis of CM. However, these differences between survival according to location were not statistically significant. In the univariate survival analysis, the only three variables that showed an impact on prognosis were the histological grade, the surrogate molecular type, and TP53 mutations. However, in the multivariate analysis, only the surrogate molecular type remained statistically significant, and the Luminal HER2- surrogate molecular type had a better prognosis.One limitation of this study is the relatively low number of cases studied. However, in spite of this and to the best of our knowledge, this is the largest series analyzing matched primary tumors and CMs reported so far. In addition, a comprehensive NGS panel of 61 genes was used in this study, including the most frequently mutated genes in BC. However, there may be additional genes with a role in progression in individual tumors that were not included in the panel. Moreover, the panel was not designed to detect CNVs or gene rearrangements, although this limitation was resolved in part by analyzing the genes most frequently amplified in BC by FISH.Further research in this field would require studies with a larger number of patients and samples and perhaps focused on patients with BC of the TN surrogated molecular type, as this seems to be not only the most likely to metastasize to the skin but also has the poorest prognosis.The development of new strategies for the management of CMs is a major clinical challenge because of the poor prognosis. To advance in this field, a better understanding of the molecular alterations involved in the metastatic process is needed. In the present study, the clinicopathological characteristics of BCs developing CM was analyzed and compared to the molecular differences between primary breast tumors and their corresponding CMs. We observed that the surrogate molecular type of BC with a greater risk to metastasize to skin was TN. A change of tumor surrogate molecular type in metastases with an impact on treatment was found in 15% of patients. In addition, half of the CM presented some additional molecular alterations with respect to the primary tumors, but a characteristic molecular pattern related to tumor progression and CM development was not observed. In this series, survival was related to the tumor surrogate molecular type. The immunohistochemical and molecular analysis of BC CM is essential for a proper treatment of the patients.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051151/s1, Table S1: Mutations and primers used for Sanger verifications. Table S2: Clinicopathological features of the 33 patients; Table S3: List of somatic mutations found in the breast cancer and cutaneous metastases subjected to NGS analysis.Conceptualization, S.G.-M., J.C. and J.P.; methodology, S.G.-M., D.P., T.C.-C. and B.P-M.; validation, S.G.-M., B.P.-M. and J.P.; formal analysis, S.G.-M. and D.P.; investigation, S.G.-M., resources, J.C., J.L.R.-P. and J.P.; data curation, S.G.-M.; writing—original draft preparation, S.G.-M.; writing—review and editing, J.C., J.P., S.G.-M., D.P., B.P.-M., T.C.-C., J.L.R.-P., G.C., M.G. and A.C.; visualization, S.G.-M., J.C. and J.P.; supervision, J.C. and J.P.; funding acquisition, J.C. and J.P.; project administration, J.C. and J.P. All authors have read and agreed to the published version of the manuscript.This study was funded by grants from the Instituto de Salud Carlos III (ISCIII) (PI19/01331) and CIBERONC (CB16/12/00316 and CB16/12/00449), co-financed by the European Development Regional Fund ‘A way to achieve Europe’ (FEDER), and by the Spanish Association Against Cancer Scientific Foundation (Grupos Coordinados Traslacionales aecc 2018).The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Hospital Ramón y Cajal (Madrid, Spain) (protocol code 30-21 19/05/2021).Informed consent was obtained from all subjects involved in the study.The data presented in this study are available on request from the corresponding author.“Contigo contra el Cáncer de la Mujer” Foundation.J.C.: Consulting/Advisor: Roche, Celgene, Cellestia, AstraZeneca, Biothera Pharmaceutical, Merus, Seattle Genetics, Daiichi Sankyo, Erytech, Athenex, Polyphor, Lilly, Servier, Merck Sharp&Dohme, GSK, Leuko, Bioasis, Clovis Oncology, Boehringer Ingelheim, Kyowa Kirin. Honoraria: Roche, Novartis, Celgene, Eisai, Pfizer, Samsung Bioepis, Lilly, Merck Sharp&Dohme, Daiichi Sankyo. Research funding to the Institution: Roche, Ariad pharmaceuticals, AstraZeneca, Baxalta GMBH/Servier Affaires, Bayer healthcare, Eisai, F.Hoffman-La Roche, Guardanth health, Merck Sharp&Dohme, Pfizer, Piqur Therapeutics, Puma C, Queen Mary University of London. Stock, patents, and intellectual property: MedSIR. Travel, accommodation, expenses: Roche, Novartis, Eisai, Pfizer, Daiichi Sankyo. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.Distribution of mutations and CNVs in this series. The figure shows paired samples corresponding to 29 paired patients, but the mutation frequencies were calculated considering the 33 paired cases with 66 samples.(a) Hematoxylin-eosin of a primary breast tumor and its corresponding cutaneous metastasis. (b) Fluorescent In-Situ Hybridization of the FGFR1 gene in the primary breast tumor (without CNV) and in the cutaneous metastasis (with polysomy). 100×.(a) Visualization in the IGV software of different ERBB2 mutations found in the primary tumor and the cutaneous metastasis in patient Pt13. (b) Orthogonal validation by the Sanger sequencing.Kaplan–Meyers graphs showing the association between overall survival and the surrogate molecular type (a), histological grade (b), and TP53 status (c). (d) Multivariate analysis showing the independent prognostic significance of the surrogate molecular type.Clinicopathological features of the 33 patients.* Local lesions refer to lesions that presented on the skin of the breast/thorax. ** Patients diagnosed at stage IV did not undergo surgery and did not have pT and pN data.Surrogate molecular type change between primary tumors and their cutaneous metastasis.Pathogenic alterations in primary tumors and their cutaneous metastasis.Additional molecular alteration in cutaneous metastases not found in primary tumors of paired cases.When not specified as an amplification, the alteration detected was a mutation.Distribution of surrogated molecular types in breast cancer with cutaneous metastasis in different series.Additional mutations in cutaneous metastases reported in different series.* Of these 8 cases, 2 had no additional molecular alterations in cutaneous metastasis.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01152.txt

@@ -0,0 +1 @@
+Stereotactic radiation therapy (SBRT) is a type of radiation therapy in which a small number of high doses of radiation are delivered to a target volume using highly accurate equipment in order to maximize cancer control while minimizing side effects on healthy tissues. SBRT’s precise role varies according to the primary location and subtype of the oligometastatic state. The purpose of this review is to clarify the role of SBRT in various cancer types and to define its position based on the oligometastatic disease state.Stereotactic body radiation therapy (SBRT) is a form of radiation therapy (RT) in which a small number of high doses of radiation are delivered to a target volume using highly sophisticated equipment. Stereotactic body radiation therapy is crucial in two cancer stages: early primary cancer and oligometastatic disease, with the goal of inducing complete cancer remission in both. This treatment method is commonly used to treat a variety of disease types. Over the years, a growing body of clinical evidence on the use of SBRT for the treatment of primary and metastatic tumors has accumulated, with efficacy and safety demonstrated in randomized clinical trials. This article will review the technical and clinical aspects of SBRT according to disease type and clinical indication.Hellman and Weichselbaum first proposed the concept of oligometastatic disease in 1995 [1]. An oligometastatic disease is a stage of the disease that is intermediate between locoregionally advanced and metastatic disease and is still treatable curatively.De novo oligometastasis, oligo-recurrence, oligo-progression and oligo-persistence are the four categories of oligometastatic disease, corresponding to the different clinical scenarios that capture the spectrum of oligometastatic disease [2,3]. The term “de novo” oligometastasis refers to newly diagnosed cancer with few metastases occurring concurrently with the primary tumor. Oligo-recurrence refers to patients who have been treated for metastatic disease and have a relapse in a few new metastatic sites. Oligo-progression refers to patients who are controlled by systemic treatments and progress only on a few metastatic sites, whereas oligo-persistence refers to patients who respond to systemic treatment but still have a few metastatic sites.It is important to note that a patient can experience dynamic transitions between oligo-recurrent, oligo-progressive, and oligo-persistent disease based on response to local and systemic therapy. For example, in a group of patients with oligometastatic prostate cancer, a median of four courses of radical local treatment were required over the course of the metastatic disease. As a result, the transition from one oligometastatic state to another is not always indicative of disease progression, but rather of a really limited oligometastatic phenotype.Ablative treatments for oligometastases must be as curative as possible and may include local surgery, radio-ablations and stereotactic body radiotherapy (SBRT).It is critical to note that SBRT has gradually been proposed as an alternative to metastasectomy and other ablative treatments. SBRT is an image-guided RT technique that delivers high doses with high precision to small target volumes in a single or small number of fractions while minimizing radiation exposure to non-targeted tissue.Before treatment can be administered, the location of the tumor or target volume must be confirmed, and immobilization devices must be used to keep the patient precisely in the same position throughout the treatment delivery [4,5]. Extracranial SBRT necessitates the use of appropriate RT devices that allow for a tight gradient of dose and a rapid decrease in dose to maximize healthy tissue organ sparing in order to guarantee maximum normal tissue spearing [6,7] (Figure 1).From a biological point of view, in addition to direct cytotoxicity, SBRT may introduce a new mechanism of radiation-induced damage, involving microvascular damage and endothelial apoptosis, resulting in microvascular disruption and death of the tissue irrigated by that vasculature [8]. Stereotactic body radiation therapy, in addition to its vasculature remodeling effect, can induce a potent “in situ” vaccination effect capable of inducing T cell infiltration as a result of high antigen load [9,10].Extracranial SBRT can be delivered to various involved organs, such as the lung, bone, liver, or adrenal glands using a variety of SBRT regimens and techniques.The multicentric “Stereotactic ablative radiotherapy versus standard of care palliative treatment in patients with oligometastatic cancers: (SABR-COMET)” phase II randomized trial now serves as a proof-of-concept for the benefit of metastasis guided SBRT in oligometastatic patients. In this trial, patients presented oligometastasis from a variety of primaries, including the colon, lung, breast, and prostate, and the sites of irradiated lesions included the lung, bone, liver, adrenal and others. Patients in the SBRT group received stereotactic radiation to all sites of metastatic disease, with the goal of achieving disease control while minimizing potential toxicities. The standard arm was the best supportive care. This trial reported a survival benefit of metastases-directed SBRT for oligometastatic patients (1–5 metastatic sites) who had their primary malignancy under control (median OS 28 months in the control group, 95% CI: 19–33 vs. 41 months in the SBRT group, 95% CI: 26-not reached; HR 0.57, 95% CI: 0.3–1.1; p = 0.09) [11]. Importantly, the overall survival (OS) benefit became larger in magnitude after a longer follow-up of 5 years, with 18% (95% CI, 6–34%) in control group vs. 42% (95% CI: 28–56%) in SBRT group (p = 0.006) [12].In terms of safety, Lehrer et al. reported a meta-analysis on 943 patients (1290 lesions) in which the estimate for late grade 3 or more adverse events was 1.2% after a median follow-up of 16.9 months. These are acceptable levels of toxicity, and validation is being performed in prospective clinical trials [13].Clinical experience and challenges in a variety of disease types are reviewed and discussed in this paper. Because the primary location is one of the most important factors influencing patient outcomes, we will discuss the findings of prospective trials that focused on specific primary sites, evaluating the outcomes of SBRT for patients with oligometastatic disease by disease type.We concentrated on the new classification of de novo, oligometastatic, oligorecurrence, and oligoprogressive disease in order to better define the strategies in the various disease settings [2,3]. Figure 2 depicts the most common types of oligometastatic disease treated with SBRT and usual doses and fractionations (Tables 1–4, Figure 2).We conducted a systematic literature review (registration number: reviewregistry1306) based on PubMed and adhered to PRISMA guidelines [14]. Two investigators (RK and FH) searched the databases independently and until 11 November 2021. The search terms were: ([oligometastasis OR oligometastases OR oligorecurrence OR oligoprogressive OR oligopersistent] AND [radiotherapy]). The results were then filtered using the following criteria: “<10 years”. We looked at prospective and retrospective trials and therapeutic interventional studies that report outcomes, such as OS, progression-free survival (PFS), or disease recurrence. Articles were excluded if they did not correspond to our review topic, if they only reported quality-of-life or if they were only about brain metastases. We prioritized prospective trials and meta-analyses to be described in the main text. We identified 972 articles that match our search terms. After applying filters, we identified 243 papers. After prioritizing, we selected 66 articles to be mentioned in the clinical results part.Prostate cancer (PCa) is the world’s second most common cancer in men and the sixth leading cause of cancer-related death. Androgen deprivation therapy (ADT) is frequently the treatment of choice for patients who have been diagnosed with metastatic or locally advanced PCa for the first time. ADT is usually combined with abiraterone, docetaxel, apalutamide, or enzalutamide in men with castration sensitive metastatic PCa. The agent chosen will be determined by the risk and clinical burden of the disease, as well as the patient’s comorbidities. Metastatic castration sensitive disease has also been divided by tumor burden; a high burden of disease has included the presence of visceral metastases, a bone-metastasis burden classified by site (beyond the axial skeleton), or a high number of lesions (more than five), or a combination of these [15,16].Patients with metastatic castration-resistant prostate cancer (mCRPC) frequently develop distant metastasis, and bony metastases can result in significant morbidity and a decline in quality of life. Individual patient data from 8820 men with mCRPC who were treated with a docetaxel-containing regimen as part of one of nine phase III trials [17] showed that OS was highest in those with lymph node-only disease and gradually declined in those with bone, lung, or liver metastases (median 31.6, 21.3, 19.4, and 13.5 months, respectively). As a result, men who have rising prostate-specific antigen (PSA) after ADT, but no evidence of macroscopic metastatic disease are classified as having non-metastatic CRPC. Multiple agents, all given in conjunction with continued ADT, have been shown in phase III trials to improve OS in men with mCRPC, and include abiraterone, enzalutamide, apalutamide, darolutamide; chemotherapy: docetaxel, cabazitaxel and immunotherapy: sipuleucel-T (in minimally symptomatic men who have a slowly progressive disease).Based on the data presented above, it is critical for PCa patients to identify intermediate stages of disease dissemination that can benefit from either improved systemic therapies or metastasis-directed therapies (MDT). Oligometastatic PCa is a broad term that encompasses at least three distinct entities, each with its own set of biological signatures and behavior. i) De novo oligometastasis refers to a distinct group of patients with PCa who have spread to limited areas prior to any definitive therapy; ii) Oligo-recurrent PCa refers to the development of limited sites of distant dissemination following primary radical prostatectomy (RP) or radiotherapy and iii) Oligo-progressive PCa refers to patients who gradually progress on less than three to five lesions despite continued systemic therapy [18] (Table 1, Figure 2).Local ablative therapy to the prostate is the most commonly used treatment in the de novo oligometastatic setting, and it has been studied in two randomized trials, which serve to reinforce the idea that treating the oligometastatic stage with ablative therapies is beneficial.The phase III HORRAD trial randomly assigned 432 men with primary metastatic PCa with bone metastases, as well as a serum PSA > 20 ng/mL, to ADT with or without external beam RT (70 Gy in 35 daily 2 Gy fractions). Two-thirds of the men had more than five bone metastases. The addition of radiation did not improve OS (the primary endpoint), but it did prolong the median time to PSA progression (median 15 vs. 12 months, HR = 0.78, 95% CI: 0.63–0.97, p = 0.02). Men with fewer than five metastases had a better chance of survival, according to an unplanned subgroup analysis, but the result was not statistically significant (HR = 0.68, 95% CI: 0.42–1.10) [19]. Similarly, 2061 men with newly diagnosed metastatic PCa were randomly assigned to ADT with or without docetaxel and with or without prostate radiation in the phase III STAMPEDE trial, (which could be either 36 Gy in six consecutive weekly fractions of 6 Gy or 55 Gy in 20 daily fractions of 2.75 Gy over four weeks). Metastatic burden was assessed at randomization using whole-body scintigraphy and computed tomography or magnetic resonance imaging staging scans, and it was classified using the CHAARTED trial definitions [20,21]. Overall survival (the primary endpoint) was not improved by prostate irradiation, but three-year failure-free survival was 32 vs. 23%, (HR = 0.76, 95% CI: 0.68–0.84, p < 0.0001). Prostate RT improved OS in men with a low metastatic burden (three-year survival 81 vs. 73%, HR = 0.68, 95% CI: 0.52–0.90) but not in those with a high metastatic burden (HR = 1.07, 95% CI: 0.90–1.28). The acute adverse effects of prostate irradiation were minor, with only 5% reporting grade 3 or 4 bladder or bowel events, compared with 1% in the control group. Approximately 1% of men who had prostate irradiation experienced late grade 3 or higher gastrointestinal toxicity, whereas none did in the control group.The pooled results of both trials, on the other hand, concluded that there was an overall improvement in biochemical PFS (HR = 0.74, 95% CI: 0.67–0.82, p < 0.00001) and failure-free-survival (HR = 0.76, 95% CI: 0.69–0.84, p < 0.00001), which translated into an approximately 10% benefit at three years for the entire cohort. In unplanned subgroup analysis of the STAMPEDE randomized trial, an OS benefit was observed in the group with three or fewer bone metastases (three-year OS 75 vs. 85%, HR = 0.64, 95% CI: 0.46–0.89) but not in those with four or more bone metastases (three-year OS 53 vs. 52%, HR = 1.12, 95% CI: 0.93–1.34). Prostate irradiation was of no benefit in patients with visceral or other metastases [22].Concerning the combination of RT with immunotherapy, Fizazi et al. explored the impact of RT (a single dose of 8 Gy,) on bone metastases (one to five metastases) followed by ipilimumab or placebo in men with mCRPC (who received docetaxel previously). The primary endpoint was OS and 799 patients were randomized. In long-term analysis the RT and immunotherapy arm presented an OS benefit of 7.4%, 6.8%, and 5.2% at 3, 4 and 5 years, respectively (HR = 0.66, 95% CI: 0.52–0.84) [23].A North American phase 2 trial is designed to test a comprehensive systemic and tumor directed therapeutic strategy for patients with newly diagnosed de novo oligometastatic PCa. Patients with newly-diagnosed M1a/b PCa and 1–5 radiographically visible metastases (excluding pelvic lymph nodes) are being treated locally with RP, six months systemic therapy (leuprolide, abiraterone acetate with prednisone, and apalutamide), metastasis-directed SBRT, and post-operative fractionated RT to the primary tumor bed if pT ≥ 3a, N1, or positive margins are present. The primary endpoint is the percentage of patients who achieve a serum PSA of <0.05 ng/mL six months after recovery of serum testosterone ≥150 ng/dL (NCT03298087, ClinicalTrials.gov) [24].The results of these trials will help us to design the next generation of clinical trials based on the concept of maximal cytoreduction.Ost et al. provided the first prospective evidence in a phase II study in which patients with oligorecurrent PCa and three extracranial metastases on choline positron emission tomography computed tomography (PET/CT) were randomly assigned to either PSA surveillance every 3 months (n = 31) or metastasis-directed therapy (MDT, surgery or SBRT) to all lesions (n = 31), with the goal of improving ADT–free survival. ADT was initiated for symptomatic or local progression, or when more than three metastases developed. After 3 years of median follow-up, the interventional group’s ADT-free survival was 21 months compared to 13 months in the control group (HR = 0.60, 80% CI: (0.40–0.90); p = 0.11). Toxicity was low, with only six patients in the MDT arm suffering from grade 1 toxicity. There was no evidence of toxicity grade 3 or higher. The authors concluded that for oligorecurrent PCa, ADT-free survival was longer with MDT than with surveillance alone, implying that MDT should be investigated further in phase III trials. Although these findings highlight the potential of MDT to delay the initiation of systemic therapy and its associated side effects, there was no statistically significant improvement in 1-year quality-of-life, possibly due to a lack of power to detect such a difference [25].An interim analysis of the phase II TRANSFORM non-randomized single institution trial looked at men who had relapsed with up to five lesions after definitive local treatment for primary PCa. The goal was to determine the proportion of patients who did not require systemic treatment after metastasis-directed SBRT. In total, 199 patients were enrolled in the study to receive fractionated SBRT (10 fractions of 5 Gy each) to all visible lesions. The authors defined the primary endpoint as the start of ADT for hormone naïve patients and the start of second-line ADT or chemotherapy for those who had prior ADT; 51.7% of patients did not require systemic therapy 2 years after SBRT (95% CI: 44.1–59.3). Over the entire follow-up period, the median length of treatment-free survival was 27.1 months (95% CI: 21.8–29.4). There was no difference in the efficacy of SBRT when treating 4–5 vs. 1–3 lesions. In 75% of patients, PSA was reduced with PSA levels felling to an undetectable level in six patients. There were no grade 3 or higher toxicities observed. The authors concluded that these interim results suggest that SBRT can be used to treat up to five synchronous PCa oligometastases to delay systemic therapies [26].Siva et al. published the findings of a single arm prospective clinical trial that investigated the safety and feasibility of single fraction SBRT for patients with oligometastatic PCa. Thirty-three consecutive patients were followed for 2 years after receiving a single dose of 20 Gy SBRT to a total of 50 lesions. Twenty patients had only bone disease, 12 had only node disease, and one had both. There was one grade 3 adverse event that was a vertebral fracture that required spinal instrumentation. The one- and two-year local control (LC) was 97% (95% CI: 91–100) and 93% (95% CI: 84–100), PFS was 58% (95% CI: 43–77) and 39% (95% CI: 25–60), respectively. The two-year freedom from ADT was 48%. The authors concluded that the SBRT approach was safe and that half of the patients in the study avoided ADT at 2 years [27].Several ongoing trials are looking into the possibility of combining local treatment with metastasis-directed RT in patients presenting oligo-recurrent PCa. This question is hypothesized in the PEACE-V trial. Patients diagnosed with Prostate-specific membrane antigen (PSMA) PET/CT detected pelvic nodal oligorecurrence (≤5 nodes) following radical local treatment will be randomized in a 1:1 ratio to arm A: MDT and 6 months of ADT, or arm B: whole pelvis RT added to MDT and 6 months of ADT. The primary endpoint is metastasis-free-survival, the estimated study completion is 31 December 2023 (NCT03569241, ClinicalTrials.gov) [28].The ORIOLE trial is a phase II randomized study evaluating the safety and efficacy of SBRT in oligometastatic hormone-sensitive PCa. Fifty-four men with oligometastatic prostate adenocarcinoma will be randomized 2:1, the primary endpoint will be PFS, the study completion date is expected mid-2023 (NCT02680587, ClinicalTrials.gov) [29].To date, no prospective trial results in patients with oligo-progressive or oligo-persistent PCa have been published.Triggiani et al. conducted a retrospective study to differentiate the results of SBRT in patients with oligorecurrent PCa from those with oligoprogressive PCa. Over 100 patients were treated with SBRT for 70 lesions, 41 of whom had oligoprogressive PCa. Progression-free-survival seemed comparable between the two study populations, the median PFS was 17.7 months in oligo-recurrent PCa and 11 months in oligo-progressive PCa. Oligoprogressive patients experienced a 2-years LC of 90.2% with no grade ≥3 toxicity. The median distant PFS was 11 months and the median second-line systemic treatment-free survival was 22 months [30].Another retrospective study looked at the outcomes of SBRT in a group of 68 patients with oligo-progressive mCRPC. Sixty-eight patients (112 lesions) were included in the study. The median time to PSA failure was 9.7 months, the time to the next intervention was 15.6 months, and the distant metastasis-free survival time was 10.8 months. When compared to a cohort of patients treated at the same institution and who only received a switch in systemic treatment but no SBRT (n = 52), SBRT was associated with a longer median time to PSA failure although the difference was not statistically significant (9.7 vs. 4.2 months, p = 0.066) [31].Main results of SBRT in oligometastatic PCa.ADT: androgenic deprivation treatment; N: number; OMD: oligometastatic disease; EBRT: external body radiotherapy; SBRT stereotactic radiotherapy; MDT: metastasis directed treatment; NA: not available; NR: not reached.Prospective trials are enrolling patients to determine the role of SBRT in the treatment of oligo-progressive PCa. The OLI-CR-P is a prospective randomized phase II study that compares the safety and efficacy of metastasis-directed SBRT to observation in patients with oligo-progressive mCRPC (NCT04141709, ClinicalTrials.gov).The TRAP trial is a multicenter, single-arm, phase II study that enrolls oligoprogressive androgen-suppressed PCa patients to evaluate the benefit of SBRT when combined with enzalutamide or abiraterone in terms of PFS (NCT0344303, ClinicalTrials.gov).To the best of our knowledge, no trial is currently underway that attempts to study SBRT in a population of oligo-persistent PCa patients, leaving an open path for studying the role of SBRT in this setting.At the time of the diagnosis, more than half of non-small cell lung cancer (NSCLC) patients are metastatic. While metastatic lung cancer has traditionally been associated with poor survival, it has become clear in recent decades that metastatic lung cancer is a heterogeneous population with varying outcomes based on the extent and location of metastatic deposits. Furthermore, advances in imaging technology and the increased use of modalities, such as brain magnetic resonance imaging (MRI) and PET/CT have allowed for more accurate staging of lung cancer patients and the detection of previously undetected metastases. Patients with oligometastatic lung cancer have been found to have better survival outcomes than patients with more widely metastatic disease, and they account for up to 25–50% of all metastatic lung cancer cases [32]. Advances in metastatic NSCLC targeted systemic therapies, such as epithelial growth factor receptor (EGFR) inhibition and immunotherapy, have improved survival outcomes, emphasizing the importance of long-term LC of metastatic deposits. Patterns of failure analyses indicate that the most likely locations of failure following first-line chemotherapy are the initially involved sites, providing additional support for MDT [33].The lung cancer group of the European Organization for research and Treatment of Cancer (EORTC) has agreed to define synchronous oligometastatic disease (sOMD) and to use it to classify patients in future clinical trials [34]. A maximum of five metastases and three affected organs were proposed in the definition, while the involvement of mediastinal lymph nodes was not considered. This definition necessitates the use of fluorodeoxyglucose (18F-FDG) PET/CT and brain imaging (preferably an MRI) to rule out the location of metastatic disease. Solitary liver metastasis should be investigated with MRI, while solitary pleural metastasis necessitates video-assisted thoracoscopy and biopsies of distant homolateral pleural locations. The minimum requirement for metastatic staging is 18F-FDG PET/CT, and histological confirmation is recommended if it affects the radiation treatment plan. When a radical disease-modifying therapy (which results in long-term disease control) is technically feasible for all tumor sites, has low toxicity, and can be offered to a patient, the term and definitions of sOMD must be used (Table 2, Figure 2 and Figure 3).Two randomized trials studied the role of local consolidative RT in patients presenting oligometastatic NSCLC [35,36]. Gomez et al. conducted a multicenter randomized phase II trial in which patients with stage IV NSCLC and ≤3 metastatic sites after first line systemic therapy (platinum doublet) were randomly assigned to MDT (SBRT or surgery) in combination with maintenance systemic therapy or maintenance systemic treatment alone. Maintenance treatment consisted of four cycles of platinum doublet or three months of EGFR or ALK inhibitors (for patients with these specific mutations) [35]. After the randomization of 49 patients, the trial was terminated early at the interim analysis due to futility. There were 25 patients in the local consolidative therapy group and 24 in the maintenance treatment group. The median PFS for local consolidative therapy was 11.9 months (90% CI: 5.7–20.9) vs. 3.9 months (90% CI: 2.3–66) for maintenance treatment (HR = 0.35, 90% CI: 0.18–0.66, p = 0.0054). In any of the groups, there was only two grade 3 toxicity with no grade 4 adverse events or treatment-related deaths. This landmark trial proved that LCT plus/maintenance therapy improved PFS compared to maintenance therapy alone in patients with ≤3 NSCLC metastases that did not progress after initial systemic therapy.Iyengar et al. conducted a single-institution randomized phase 2 study comparing maintenance chemotherapy alone to SBRT followed by maintenance chemotherapy for patients with limited metastatic NSCLC [36]. Patients had to have tumors that did not have EGFR- or ALK-targetable mutations but achieved a partial response or stable disease after induction chemotherapy. Patients were irradiated on the primary site as well as up to five metastatic sites. The primary endpoint was PFS. A total of 29 patients were enrolled in the study; 14 were allocated to the SBRT-plus-maintenance chemotherapy arm, and 15 to the maintenance chemotherapy–alone arm. The trial was terminated early after an interim analysis revealed a significant improvement in PFS in the SBRT-plus-maintenance chemotherapy arm of 9.7 months vs. 3.5 months in the maintenance chemotherapy–alone arm (p = 0.01). The toxicity was comparable in both arms, with two grade 3 toxicities in the maintenance arm alone and four grade 3 toxicities in the SBRT-plus-maintenance arm.A meta-analysis of 21 studies looked into the addition of local thoracic RT to standard-of-care systemic treatment in patients with sOMD NSCLC [37]. The median OS and PFS were 20.4 and 12 months, respectively. The pooled 1-2-3 and 5-year OS rates were 70.3%, 43.5%, 29.3% and 20.2%, respectively. The addition of thoracic RT improved OS (HR = 0.44, 95% CI: 0.32–0.6; p < 0.001). Similarly, adding RT to the primary tumor improved PFS (HR = 0.42, 95% CI: 0.33–0.55; p < 0.001).These trials shared the same belief in aggressive local treatment for patients with a low metastatic burden. However, additional research should be conducted to confirm this data ideally in phase III randomized studies. Gomez et al. and Iyengar et al. opened the path for randomized trials for evaluating the impact of localized treatment for oligometastatic NSCLC. Furthermore, because these trials were conducted prior to the immuno-oncology era, the same questions should be investigated with the inclusion of novel systemic agents, such as immunotherapy.Bauml et al., conducted a single arm phase 2 study that enrolled 51 patients with sOMD and metachronous NSCLC (less than four metastatic lesions) after first line chemotherapy to evaluate the effect of SBRT or surgery following 4 weeks after treatment with an anti-PD-L1 immune checkpoint inhibitor (pembrolizumab, 200 mg every 21 days). Patients were not selected based on PD-L1 status but 34% had results positive for PD-L1 (≥1%) and 52% had CD8 T-cell infiltration of greater than 2.5%. After a median follow-up of 25 months, there was a statistically significant improvement in median PFS from historical control from 6.6 months to 19.1 months (95% CI: 9.4–28.7 months; p = 0.005). Progression was local only (at a site of a prior SBRT) in two patients, systemic only (outside SBRT volume) in 15 patients, and both in six patients [38].This information led Theelen et al. [39] to perform the PEMBRO-RT trial, a randomized phase II study that included 92 patients with advanced stage NSCLC. The goal of this trial was to assess whether the addition of SBRT to a single tumor lesion prior to pembrolizumab enhances response in stage IV NSCLC patients. Ninety-two patients were randomly assigned to receive either pembrolizumab (200 mg/kg every 3 weeks) administered alone or after SBRT to a single tumor lesion until progression, unacceptable toxicity, or a maximum of 24 months. In the SBRT arm, the first pembrolizumab dose was given ≤7 days after completion of SBRT, consisting of three doses of 8 Gy delivered on alternate days to a single tumor site. The 3-month response rate was 18% in the control arm vs. 36% in the SBRT arm (p = 0.07). Median OS was 7.6 months in pembrolizumab alone vs. 15.9 months in the SBRT arm, (HR = 0.66; 95% CI: 0.37–1.18; p = 0.16). A significant improvement (64% vs. 40%; p = 0.04) was observed in the disease control rate at 12 weeks in the SBRT arm. The 14–18 CHESS-ETOP trial (NCT03965468) is a multicenter single arm phase II study designed to evaluate the efficacy of immunotherapy, chemotherapy and SBRT to metastases followed by definitive surgery or RT to the primary tumor, in patients with sOMD NSCLC. The primary endpoint is PFS. Patients will benefit from definitive primary treatment (surgery or curative radio-chemotherapy), SBRT to all oligometastatic sites and maintenance durvalumab for a maximum of 1 year until progression. The first patient was treated in November 2019, 47 patients are planned to be enrolled. If the results are positive, it could serve as yet another argument in favor of this approach in the context of combinatorial chemo-immunotherapy.A prospective phase II non-randomized trial involving 25 patients with oligoprogressive disease looked into the role of SBRT in patients who presented extra-cranial oligometastasis and EGFR mutated NSCLC receiving erlotinib. SBRT to progressive sites with tyrosine-kinase inhibitors (TKI) maintenance resulted in a 6 months median PFS and 29 months median OS (95% CI: 21.7–36.3) [40]. Similarly, in EGFR mutant NSCLC, when local SBRT was added to oligo-progressive lesions in a multi-institutional phase II trial, median PFS and OS were significantly greater of 15 months, and 20 months, respectively, than historical controls receiving systemic drugs alone. However, two patients had grade 3 toxicities related to SBRT (pneumonitis and back pain) [41].Weickhardt et al. reported the retrospective results of 24 patients with oligo-progressive NSCLC who after initial progression on targeted therapy continued to receive the same drug in conjunction with SBRT. The disease control benefit was 6.2 months compared to the continuation of the drug alone. [42].Chan et al. identified 25 patients who received SBRT for three or fewer oligo-progressive lesions and continued their systemic treatment with oral TKI. The results were compared to those of a group of patients with oligo-progressive NSCLC who received a systemic line switch. The study concluded that metastasis-directed SBRT provided an OS advantage of 10 months (28.2 vs. 14.7 months) compared to a switch of systemic therapy. Only one patient presented a grade ≥3 toxicity after RT [43]. Another retrospective study in a similar population of 46 patients with oligo-progressive NSCLC with druggable mutations found that after local SBRT and continuation of the TKI, the PFS was 7 months with no grade ≥4 [44].Main results of SBRT in oligometastatic NSCLC.N: number; OMD: oligometastatic disease; SBRT stereotactic radiotherapy; MDT: metastasis directed treatment; NA: not available; NR: not reached.Several prospective trials are currently enrolling patients with oligo-progressive NSCLC in order to refine the effect of SBRT and its impact on OS, PFS and quality of life.The STOP trial is a phase II trial in which patients with oligo-progressive NSCLC are randomly assigned to receive standard of care systemic therapy plus SBRT to all sites of progressive disease or standard-of-care systemic therapy alone. A total of 90 participants are expected to be enrolled until June 2022 (NCT02756793, ClinicalTrials.gov).HALT is a multicenter phase II/III trial that aims to enroll patients with mutation-positive advanced NSCLC who are receiving targeted therapy and have oligo-progressive disease. A total of 110 patients are expected to be enrolled and randomly assigned 2:1 to either SBRT or no SBRT (NCT03256981, ClinicalTrials.gov).SUPPRESS-NSCLC is a phase II trial that randomly assigns patients with NSCLC who have evidence of oligo-progressive disease while on immune-checkpoint inhibitor or tyrosine kinase inhibitor regimens to SBRT plus current systemic medication or standard of care (NCT04405401, ClinicalTrials.gov).A phase II non-randomized prospective study (ATOM) tried to assess the efficacy of SBRT to oligo persistent lesion after 3 months of EGFR TKI treatment. Eighteen patients with ≤4 lesions were enrolled from 2014 to 2017; recruitment was stopped before the planned number of 34 because of slow accrual. The 1-year PFS was 68.8%, and there was no grade 3 or more toxicity [45].Unfortunately, to the best of our knowledge, no prospective trials are being conducted in the sub-population of patients with oligo-recurrent NSCLC. This population still needs to be investigated in future clinical trials.Breast cancer (BC) is the most common cancer in women worldwide, as well as the second leading cause of cancer-related death. The majority of BC deaths are the result of a distant recurrence or metastatic disease. De novo oligometastatic BC accounts for approximately 6% of all cases of metastatic BC, and 20–30% of all early-stage BC will relapse on a distant site. Given the long natural history of some metastatic BC, particularly those with hormone receptor positive disease and bone-only metastases, it appears ideal to treat all oligometastases with local therapy [46,47].Similarly, for patients with de novo oligometastatic disease (untreated primary tumor plus limited metastases) it is uncertain whether surgery to the primary with or without adjuvant local RT is better than systemic therapy alone.There are also several treatment systemic options for oligometastatic BC, like chemotherapy, target therapy, immunotherapy, or a combination of these approaches which makes the oligoprogressive scenario an ideal opportunity for incorporating metastasis-directed SBRT [48,49,50,51,52,53,54].A study of 361 “all comers” extracranial oligometastatic cancer patients treated with SBRT sought to identify prognostic pretreatment factors to identify which patients may benefit the most from MDT [55]. Median OS was 47.1 months, with BC patients having significantly longer OS than colorectal, gastrointestinal, NSCLC, sarcoma, and other primary tumor types (Table 3, Figure 2). Locoregional treatments for primary breast tumors led to incongruent results that did not clearly identify a population that would benefit from breast surgery [56,57,58].From 2005 to 2013, 716 women with de novo metastatic BC were randomized to receive locoregional treatment with surgery and adjuvant radiation at Tata Memorial Hospital. Median OS was not different between the two groups (19.2 vs. 20.5 months, p = 0.79). However, this study’s systemic therapy was criticized (e.g., limited taxane use; 92% of patients with HER2 positive disease did not receive anti-HER2 therapy) [59].Between 2007 and 2012, a Turkish study called MF07-01 randomized 274 treatment-naive patients with stage IV BC to receive locoregional treatment followed by systemic therapy or systemic therapy alone. The 3-year survival rates were similar in both groups (60 vs. 51%, p = 0.1). The locoregional group had a median survival of 46 months compared to 37 months in the systemic therapy group (HR = 0.66, 95% CI: 0.49–0.88, p = 0.005). Patients with positive estrogen and progesterone receptors (ER+/PR+), HER2 negative, younger than 55 years of age, and solitary bone-only metastases benefited from local therapy [60].Main results of SBRT in oligometastatic BC.N: number; OMD: oligometastatic disease; FFDM: freedom from widespread distant metastasis; EBRT: external body radiotherapy; SBRT stereotactic radiotherapy; MDT: metastasis directed treatment; NA: not available; NR: not reached.The place of locoregional treatment in stage IV disease should thus be further investigated.Several studies are underway in order to investigate the role of metastasis-directed SBRT in the de novo oligometastatic BC.The NRG oncology group designed a phase II/III trial that will study the impact of metastases guided RT in patients with de novo oligometastatic BC (NRG-BR002, ClinicalTrials.gov). Three hundred sixty patients will be randomized between the continuation of their current planned systemic therapy at the discretion of the treating physician (Arm 1) or the Arm 1 treatment with the addition of SBRT to metastatic sites. The primary endpoints will be PFS and OS, the study was suspended in September 2021 for interim analysis (NCT02364557, ClinicalTrials.gov).STEREO-SEIN is a large prospective trial (n = 280) aiming at PFS improvement, the experimental arm will receive SBRT to all metastases and the beginning of systemic treatment will be administered 2 to7 days after SBRT completion, while the active comparator will not get SBRT (NCT02089100, ClinicalTrials.gov).CLEAR is a multicenter, single-arm, phase 2 trial that will investigate the role of local treatment in addition to endocrine therapy in ER-positive/HER2-negative oligo-metastatic de novo BC. One hundred and ten patients are expected to be enrolled until mid-2025 (NCT 03750396, ClinicalTrials.gov). Another phase II/III trial will test whether treating BC metastases with surgery or high-dose radiation improves survival (OS and PFS). Until the end of 2022, 360 participants will be randomized to MDT (SBRT or surgery) or continuation of systemic therapy (NCT02364557, ClinicalTrials.gov).Similarly, to the de novo setting, evidence for other subtypes of oligometastasis is lacking in BC, as prospective trials are uncommon and the patient population is not well defined.Miyata et al. investigated the place of RT in a group of 21 patients treated for an oligo-recurrent BC relapse. The second oligometastatic relapse occurred after a median of 24 months, and the OS was 41 months. Toxicities were mild with only one grade 3 acute toxicity. The authors came to the conclusion that RT directed to oligo-metastasis could differ in time to a new distant recurrence [61].A prospective phase II multicentric trial was designed to determine if administering MDT to all metastatic sites could improve the PFS in patients with oligometastatic BC. Patients presented with BC with up to five metastatic sites, no brain metastases, and they presented a non-treated primary tumor. SBRT technique or fractionated intensity-modulated RT (IMRT) were permitted. Fifty-four patients with 92 metastatic lesions were included in the study. The one-year and two-year PFS rates were 75% and 53%, respectively. The two-year LC and OS rates were 97% and 95%, respectively. RT was well tolerated with no evidence of grade 3 toxicity. The authors came to the conclusion that radical radiation therapy to all metastatic sites should be used in patients with oligometastatic BC [62].Milano et al. published an update on the results of a phase II nonrandomized prospective trial involving forty-eight women with 1–5 extracranial BC oligometastases who received SBRT to all radiographically visible sites of disease. After SBRT, the 5- and 10-year OS rates for patients who suffered from bone only metastasis were 83% and 75%, respectively, while for patients with visceral disease the 5- and 10-years OS rates were 31% and 17%, (p = 0.002). The tumor burden, the number of oligometastatic lesions and the presence of visceral metastasis were significant factors of freedom from widespread metastasis. This emphasizes that patients with BC with oligometastatic disease treated with SBRT can have a positive outcome but that this depends on the volume and the number of lesions as well as their location (visceral vs. bone) [63].In the same line, David et al. reported the results of a single institution prospective trial on single fraction SBRT for patients with bone only oligometastatic BC. Each patient received 20 Gy in one fraction to each metastasis (1–3 lesions) [64]. The two-year LC was 100% and the PFS was 67%. SBRT was safe and effective in this cohort, with two-thirds of the patients disease-free after two years. No patients experienced a grade 3 or more toxicity [64].The phase II randomized CURB trial has evaluated the benefit of SBRT in metastatic NSCLC and BC. However, the 12-week PFS benefit was only found in NSCLC patients, and not in BC patients. It is necessary to continue to investigate the role of metastasis guided SBRT in BC and why there may be a difference in benefit between cohorts depending on the primary [65].About 16% of renal cell carcinoma (RCC) patients present with locally advanced or de novo metastatic disease at diagnosis for which surgery is not feasible [66]. The natural history of advanced or metastatic RCC varies from months to years depending on clinical, pathologic, laboratory, and radiographic features [67].Depending on the extent of disease, sites of involvement, and patient-specific factors, systemic therapy (immunotherapy, molecularly targeted agents), surgery, and RT may all play a role.Systemic therapy is the cornerstone of treatment for de novo metastatic RCC, and new guidelines adapted from the European Association of Urology (EAU) and European Society of Medical Oncology (ESMO) and based on the International Metastatic RCC Database Consortium (IMDC) risk classification agreed that different combinations of immunotherapy and anti-angiogenic therapy must be offered upfront to newly diagnosed metastatic RCC [68] (Table 1).Over the past decade, evidence suggests that in oligometastatic RCC aggressive local therapy could improve outcomes. RCC was historically known to be radio-resistant to conventional RT; however, important clinical responses have been observed in patients treated with SBRT which serves to reinforce the concept that RCC may not be as radioresistant as previously thought [69]. For instance, a meta-analysis of 28 studies looked at the role of SBRT in the treatment of oligometastatic RCC [70]. There were 679 patients with a total of 1159 extracranial lesions. The median treatment volume was 59.7 cc, the 1-year LC rate was 89.1% and the 1-year survival rate was 86.8%. For extracranial disease, the incidence of any grade 3–4 toxicity was 0.7%.Moreover, in the metastatic setting in patients receiving ≤2 prior anti-angiogenic therapies, the non-randomized phase II NIVES study tested the combination of SBRT (3 fractions of 10 Gy each) concomitant to immunotherapy (nivolumab, anti-PD1, 240 mg every 14 days for 6 months). Sixty-nine patients were enrolled. The overall response rate (ORR) was 17% and the disease control rate was 55%. The median PFS was 5.6 months (95% CI, 2.9–7.1) and median OS 20 months (95% CI, 17-not reached). After 1.5 years of follow-up, 23 patients died. The median duration of response was 14 months. No new safety concerns arose. [71]. In the same context, the RADVAX trial investigated if patients with metastatic RCC receiving nivolumab and ipilimumab (anti-CTLA-4 antibody) benefited from the addition of SBRT (five fractions of 10 Gy each) to 1–2 metastatic sites administered between the first and second dose of immunotherapy. The ORR in the 25 enrolled patients was 56%, while two grade 2 toxicities were observed [72].In the UT Southwestern phase II single arm study, 47 patients with de novo oligometastatic RCC, were treated with SBRT on 88 extracranial lesions prior to starting systemic therapy [73,74]. The LC rate was 91.5% at two years, with no grade 3 toxicity. The median time to start systemic therapy was 15.2 months and the percentage of patients with no metachronous illness at 1 year improved significantly. The same group of investigators is now planning a prospective trial that will investigate the role of SBRT in this particular population of de novo oligometastatic RCC. The primary endpoint is the time to start systemic therapy, 23 patients are expected to be enrolled until the end of 2023 (NCT02956798, ClinicalTrials.gov).These studies show that SBRT is safe and effective in RCC, and its use should be tailored to specific situations, such as when a patient is oligometastatic and the treating physician wants to postpone a change in systemic therapy, or when a patient cannot receive either anti-angiogenic therapy, immunotherapy, or a combination of the two.In some cases, patients progressing on immunotherapy required SBRT due to oligoprogression; in this case, SBRT should not be used to look for an abscopal effect, which is a rare event in itself, but rather to reduce metastatic tumor burden.Similarly, SBRT should be considered as an alternative to surgery or invasive MDT in patients with metastatic RCC and untreated primary tumors.Santini et al. retrospectively enrolled 55 patients who experienced disease oligo-progression after at least 6 months from the beginning of first-line therapy and were treated with MDT. The median time to the first relapse after MDT was 14 months. Patients who received the same therapy after SBRT treatment on a site of progression had significantly longer OS (from the time of first oligo-progression) than those who switched therapies (39 vs. 11 months, p = 0.014) [75].Prospective trials are being developed to assess the role of SBRT in oligo-progressive RCC patients. GETUG-StORM is a multicenter phase II prospective trial that will investigate the efficacy of SBRT in prolonging PFS in patients with oligo-metastatic RCC and in which proportion it can delay the initiation of systemic therapy. Patients will be randomized 2:1, 114 patients are expected to be included until the end of 2023 (NCT02956798, ClinicalTrials.gov). Another single institution trial will enroll patients in the same setting with a similar endpoint (patients under first line Sunitinib), 38 participants will be enrolled (NCT02019576, ClinicalTrials.gov).A single arm phase II study looked at the feasibility of SBRT to all metastatic sites that remained after a first line of systemic treatment for RCC (TKI or immunotherapy) and in patients with de novo oligometastatic disease. Patients could be treated on up to five lesions, and they had to stop their systemic treatment at least one month before SBRT. Thirty patients were enrolled, the median PFS was 22.7 months and no grade 3 complications were observed. The authors concluded that sequential RT could defer systemic therapy and allow for systemic therapy breaks in patients with oligo-persistent RCC [76] (Table 4).Colorectal cancer is the third most common cancer affecting both males and females in Europe. Approximately 20 to 25% of newly diagnosed colon cancers are metastatic at presentation (synchronous metastasis). Others may develop metastatic disease after potentially curative treatment of localized disease. The most common distant metastatic sites are the liver, lungs, lymph nodes, and peritoneum.Despite significant advances in systemic chemotherapy that have increased median survival from less than one year in the single-agent fluoropyrimidine era to more than 30 months, fewer than 20% of those treated with chemotherapy alone are still alive at five years, and only a few are disease-free unless metastasis resection or ablation is timely offered [77]. Surgery, on the other hand, offers a potentially curative option for selected patients with limited metastatic disease, most commonly in the liver and lung. Metastasectomy can result in long-term survival in up to 50% of cases, and an aggressive surgical approach to both the primary and metastatic sites is required in conjunction with systemic chemotherapy. However, even after complete resection of metastases, the majority of patients who survive five years have the active disease; only about 20 to 30% are free of recurrence long-term and may be cured.To classify patients with oligometastatic CRC with liver metastasis, Pitroda et al. developed an integrated molecular classification based on the analysis of 134 patients that benefited from liver metastasectomy. Three subtypes were defined (low risk, intermediate risk, and high risk patients), with 10-year OS rates of 94%, 45%, and 19%, respectively [78]. Poor-prognosis subtypes have VEGFA amplification in conjunction with stromal, mesenchymal, and angiogenic signatures (Subtype 3 stromal), or exclusive NOTCH1 and PIK3C2B mutations with E2F/MYC activation (Subtype 1 canonical) [78]. To the best of our knowledge, no prospective or retrospective studies have demonstrated the outcome of patients based on the subset of oligometastatic disease they have.More pertinent data can be found in the meta-analysis presented here.Patients with oligometastasis in the liver are typically offered surgery. SBRT is usually offered to treat inoperable lesions. A meta-analysis of 18 studies involving 656 patients with oligometastatic CRC and treated by SBRT on the liver found a one-year OS of 67.2% (95% CI: 42.1–92.2) and a two-year OS of 56.5% (95% CI: 36.7–76.2), respectively. The median PFS and OS were 11.5 and 31.5 months, respectively. The pooled one-year LC was 67.0% (95% CI: 43.8–90.2), and the pooled two-year LC was 59.3% (95% CI: 37.2–81.5). Mild-moderate and severe liver toxicity were 30.7% and 8.7%, respectively. SBRT for liver oligometastases is an effective treatment option for patients with advanced CRC, with encouraging LC and survival [79] (Table 1).The largest multicenter retrospective study on the topic of lung oligometastasis from CRC primary showed a 75.4% LC rate at 2 years after analyzing the outcomes of 1033 lesions. LC was significantly improved when the lesion was less than 2 cm in diameter and the treatment dose was at least 125 Gy BED (biological equivalent dose). A meta-analysis attempted to summarize the outcomes of SBRT for CRC lung oligometastases. Eighteen retrospective studies with a total of 1920 patients found that the LC rate in patients with CRC pulmonary oligometastases was significantly lower than in patients with other cancers (odds ratio 3.10, p = 0.00001) (Table 1, Figure 2) [80].After publishing encouraging results from the SABR-COMET phase II trial, the authors concluded the necessity to confirm their results through a larger phase III study [11]. COMET-10 is a prospective phase III trial that will include patients presenting 4–10 metastatic lesions [81]. One hundred and fifty-nine patients will be randomized to receive standard-of-care palliative-intent treatments (control arm), or standard-of-care treatment and SBRT to all sites of known disease (SBRT arm). This study will provide an assessment of the impact of SBRT on clinical outcomes and quality-of-life, to determine if long-term survival can be achieved for selected patients with 4–10 oligometastatic lesions. The trial is registered on ClinicalTrials.gov, NCT03721341, and started on 22 February 2019, the completion date is estimated for January 2029.SABR-5 is a population-based phase II trial of SBRT for up to five oligometastases. In this non-randomized phase II trial, all participants will receive experimental SBRT treatment to all sites of newly diagnosed or progressing oligometastatic disease. Two hundred patients will be enrolled (expected primary completion 2025) to assess toxicity associated with this experimental treatment and measure late grade 4 toxicity [82]. EORTC is also conducting a phase III superiority study (OligoRARE, ClinicalTrials.gov) comparing the effect of adding SBRT to the standard of care treatment on OS in patients with rare oligometastatic cancers. Patients will be randomized in a 1:1 ratio between current standard of care treatment vs. standard of care treatment + SBRT to all sites of known metastatic disease. The primary objective of this trial is to assess if the addition of SBRT to the standard of care treatment improves OS as compared to standard of care treatment alone in patients with rare oligometastatic cancers. Two hundred patients will be recruited from June 2021 to August 2028. Expected results in 2030 should assess the results of SABR-COMET cohorts.The term oligometastatic disease originated in the 1990s, during a period of less precise radiation technology. With the possibility of a patient having multiple metastases that can be treated concurrently or sequentially with SBRT, the concept of oligometastases should be reconsidered.For example, in neuro-oncology, stereotactic radiosurgery is now possible to treat more than 10 to 15 individual lesions concurrently without experiencing significant toxicities [83,84].What we do not know is the maximum number of body metastases that can be treated safely in this manner. Numerous constraints currently prevent the widespread use of multisite SBRT. When high-dose RT is delivered to multiple targets, the first constraint is a lack of knowledge regarding appropriate cumulative dose constraints. While current dose-volume histogram measurements and constraints provide excellent detail on doses to critical organs, scatter dose (e.g., volumes receiving 5–10 Gy) may become significant issues in plans with multiple targets, such as at the level of hematological toxicity.Multi-site irradiation should be evaluated from a feasibility standpoint because it requires repeated measurements, contouring, as well as longer treatment times. Numerous tools and techniques are being developed to assist in the development of a suitable workflow. New technologies in development (e.g., EthosTM, Varian Medical System, for adaptative radiotherapy using artificial intelligence, or RayIntelligence®, RaySearch Laboratories AB, using deep learning analytics system) may enable more rapid treatment of multiple sites, thereby, reducing planning and treatment time and mitigating the effects of target motion and uncertainty [85,86]. Auto-contouring, auto-planning, and auto-quality assurance tools, for example, can significantly reduce the time to prepare and deliver treatments [87].As a result, we believe that the unanswered research question in our field is now whether SBRT is safe and feasible when administered concurrently or sequentially to multiple (>5) metastatic sites. If proven to be safe, multi-site SBRT would be an ideal complement to the growing effectiveness of systemic treatments.Stereotactic body radiation therapy is a type of radiation therapy that involves delivering a small number of high doses of radiation to a target volume; this technique has quickly gained popularity due to its excellent tolerability and high loco-regional control rates that approach 90%. The routine use of SBRT necessitates careful consideration of organ motion, and adaptive technology is constantly evolving to make these treatments more precise, resulting in fewer and fewer side effects.To date, evidence for treating oligometastasis is based on the proof-of-concept results of the SABR- COMET randomized trial, which demonstrated a significant increase in OS with the use of SBRT compared to best supportive care.We believe that future studies should focus on randomizing patients based on disease type and stratifying patients based on the number of metastases, as well as disease location (bone vs. visceral), as these are distinct prognostic factors that should be considered.New guidelines have established the use of specific nomenclature: oligorecurrence, oligoprogression, and oligopersistence taking into account whether oligometastatic disease is diagnosed during a treatment-free interval or during active systemic therapy, as well as whether an oligometastatic lesion is progressing on current imaging. This oligometastatic disease classification and nomenclature should be evaluated in clinical trials in the future because each disease subtype and oligometastatic state may have different outcomes.SBRT could then be used in a dynamic conception of oligometastatic disease, and if proven to be safe, multi-site SBRT would be an ideal complement to the growing effectiveness of systemic treatments. Efforts should also be made to obtain paired pre-post treatment tumor biopsies in order to determine which patients benefit the most from SBRT.The concept of oligometastatic disease, as well as the implementation of SBRT, should not be viewed in isolation from the current systemic treatment that a given disease type requires, and its incorporation into standard cancer patient management will require not only prospective randomized trials but synergistic multidisciplinary teams capable of evaluating patients on a case-by-case basis and deciding when and how to incorporate SBRT in a given clinical scenario.R.K. and F.G.H.: conception, design, acquisition of data, analysis, and interpretation of data, drafted the work and substantively revised it; both authors approved the submitted version, E.M.: figures design, drafted the work and substantively revised it; approved the submitted version, L.S. and J.B.: substantively revised the manuscript and approved the submitted version. All authors have read and agreed to the published version of the manuscript.This research received no external funding.Not applicable.Not applicable.R.K., E.M. and L.S. declare no conflict of interest. J.B. reports advisory role for Roche, BMS, MSD, Astra-Zeneca, Debiopharm, Nanobiotix, Merck and Mevion; F.G.H. reports grants from Prostate Cancer Foundation, Bristol-Myers-Squibb, Accuray Inc, Bioprotect, and non-financial support from Roche ImFlame cooperative group, European Organization for Research and Treatment of Cancer (EORTC) chairman Gynecology Cancer Group. F.H. has received honoraria for consultations from Johnson and Johnson.Steps for planning a stereotactic body radiotherapy (SBRT) treatment. (a) Consultation with the patient to address the treatment goals and potential side effects, (b) Placement of fiducial markers for tumor tracking purposes all along the treatment, (c) 3D imaging CT scan that provides tumor’s precise location, (d) Treatment dosimetry planning provides dosage level and positioning of radiation beams, (e) Treatment is delivered with sophisticated machines that allow for daily imaging to ensure proper tumor positioning, (f) Potential increase in OS observed in some studies.SBRT dose and fractionation depending on metastatic sites.Dosimetry of a 55 Gy in five fractions of 11 Gy lung SBRT for pulmonary metastasis from lung cancer primary. (a) axial view, (b) coronal view. Gross tumor volume (GTV) outlined in red, planning target volume (PTV) outlined in blue and fiducial markers outlined in green.Main results of SBRT in oligometastatic RCC.N: number; OMD: oligometastatic disease; FST freedom from systemic therapy; EBRT: external body radiotherapy; SBRT stereotactic radiotherapy; MDT: metastasis directed treatment; NA: not available; NR: not reached.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01153.txt

@@ -0,0 +1 @@
+Membership of the characterization and molecular subtyping of interval cancers in colorectal cancer screening research group (MSIC-SC) is provided in the Acknowledgments.Hemoglobin degradation can be affected by ambient temperature and humidity. How this modifies the advanced neoplasia detection rate and interval cancer rate remains understudied. We conducted a retrospective study and analyzed the impact of ambient temperature and humidity on the fecal immunochemical test (FIT) positivity rate, detection rate for advanced neoplasia, and interval colorectal cancer (CRC). The results of our study indicated that at >24 °C, the positivity rate was lower, whereas the detection rate of the FIT for advanced neoplasia and the interval cancer detection rate were not affected, probably because we have adopted measures to minimize the impact of ambient temperature on FIT sensitivity. Humidity did not affect FIT sensitivity. The results emphasize the importance of organizational efforts on the procedures along the screening process (such as the cold chain) to minimize the effect of seasonal variations in temperature on the positivity rate.Exposure of the fecal immunochemical test (FIT) to different ambient temperatures and humidity is unavoidable in population-based screening programs in Southern European countries, and it could lead to a decrease in target colorectal lesions. The objective was to evaluate the effect of ambient temperature and humidity on the FIT sensitivity in a population-based screening program for colorectal cancer (CRC) using an ecological design. The retrospective cohort included individuals aged 50–69 years who participated in CRC screening (Barcelona) from 2010–2015, and were followed until 2017 to identify interval CRCs. The positivity rate, and detection rates for advanced polyps and CRC were compared according to ambient temperature, humidity, and quarters of the year. A positive FIT was defined as the detection of ≥20 μg Hb/g in feces. The monthly ambient temperature and humidity were recorded on the day that the FIT was performed. In total, 92,273 FIT results from 53,860 participants were analyzed. The FIT positivity rate was lower at >24 °C than at ≤24 °C (p = 0.005) but was not affected by humidity. The temperature’s impact on positivity did not lead to a decrease in the FIT detection rate for advanced neoplasia or the interval cancer detection rate in a program where the samples were refrigerated until the analysis and screening invitations were discontinued in July and August.Colorectal cancer (CRC) screening based on fecal occult blood followed by a diagnostic colonoscopy reduces CRC mortality [1]. The fecal immunochemical test (FIT) is the preferred screening test for CRC in most organized screening programs [2,3]. The interval cancer rate is strongly correlated with the sensitivity of the FIT and reflects the quality of the screening program. In a population-screening program, the FIT has a sensitivity of 29% and a specificity of 97% for detecting advanced adenoma, and a sensitivity of 86% and a specificity of 85% for detecting CRC [4,5]. Studies of FIT stability [6,7,8,9,10] and the FIT manufacturers’ specifications suggest shorter intervals between collection and testing as continued exposure to ambient temperature decreases the test performance of the FIT. Several studies, mainly with a cross-sectional design, have examined the FIT’s performance when samples are returned during warm months with inconsistent results. A few of them suggest that the positivity rate of the FIT is reduced with high ambient temperatures [11,12,13,14,15], but some other investigations have reported no significant variations in positivity rates according to seasonal temperatures [16,17,18,19]. Moreover, Adam et al. [20] suggested that humidity could also be important in maintaining the performance of FIT in an experimental study. Only Park et al. [17] has analyzed the effect of temperature and humidity in a CRC screening setting. They reported that high temperature and high humidity decreased FIT’s positivity rate.The existing literature on the impact of ambient temperature on FIT and advanced neoplasia detection in a population-based screening program is scarce [13,14,15,21]. Only two studies [13,14] have evaluated the risk of interval cancer, suggesting, though with caution, that interval cancers were more frequently detected in the summer and autumn seasons. In fact, the US Multi-Society Task Force [22] states insufficient evidence to recommend against distributing or mailing FITs when ambient temperatures are above a certain level.This study aimed to evaluate the effect of ambient temperature and humidity, both independently and in combination, on the FIT’s sensitivity using a population-based screening cohort. The positivity rate and FIT’s sensitivity for advanced polyps and CRC were assessed according to ambient humidity and temperature.A biennial screening program for CRC, which is free of charge, is managed by the Catalan Institute of Oncology. We used the FIT with a cut-off of ≥20 μg Hb/g feces (100 ng/mL) (OC-Sensor, Eiken Chemical Co., Tokyo, Japan) as the screening test. The target population of the current study (n = approximately 495,000) includes men and women aged 50–69 years from the Northern and Southern metropolitan areas of Barcelona (Catalonia, Spain). The exclusion criteria for the screening program were gastrointestinal symptoms, advanced polyps’ history, hereditary CRC syndromes, familial or personal history of CRC, inflammatory bowel disease, colonoscopy in the previous five years or a FIT within the last two years, terminal illness, and severely disabling conditions. The description of our screening program and its quality indicators has been described previously [23,24].All eligible subjects received an invitation letter to collect a FIT kit at any nearby community pharmacy. We sent the invitations according to the primary healthcare areas (territorial divisions through which primary health care services are organized). Individuals in a given geographic area were assigned to a primary care team and an endoscopic unit for the diagnostic colonoscopy. Once the sample was collected, subjects were instructed to store the sample in the refrigerator and take it to a nearby pharmacy as soon as possible. The sample was exposed to ambient temperatures during transport from home to pharmacy. Subsequently, it was transported refrigerated to a central testing center by a courier. The distribution was daily or at least three times a week. Samples were rejected if there was a delay of ≥12 days from the time of collection to the laboratory, and patients were sent a new kit to collect a new sample. Among participants who completed the FIT, the median time of the return of the kit was three days. Table S1 shows the conservation temperature, maximum days for each step of the process, and if the temperature was controlled. During the summer holidays (July and August), the program usually does not send invitations. However, the participation rate is similar in the third quarter (July–September) and the first quarter (January–March).For the study, we selected individuals with a conclusive FIT result from October 2009 to December 2015. The minimum follow-up was 24 months. Endoscopic findings were classified according to the European guidelines for quality assurance in CRC [25]. Low-risk lesions were defined as one or two tubular adenomas measuring <10 mm with low-grade dysplasia; intermediate-risk lesions were defined as three or four tubular adenomas measuring <10 mm with low-grade dysplasia, or as one to four adenomas measuring 10–19 mm with low-grade dysplasia or at least one with tubulovillous/villous or carcinoma in situ or with high-grade dysplasia; high-risk lesions were defined as ≥5 adenomas or ≥1 adenoma measuring ≥20 mm. Subjects were classified according to their most advanced lesion. The CRCs were divided into (1) screen-detected cancer: cancer detected after a positive FIT, and (2) interval cancer: cancer diagnosed after a negative FIT result and before the next screening invitation (≤24 months). We defined a post-colonoscopy CRC as a CRC diagnosed after a diagnostic colonoscopy in which no CRC was detected before the next recommended exam date. As per [25,26,27], positive FITs without subsequent colonoscopy or with a low-quality colonoscopy (incomplete or inadequate bowel preparation), post-colonoscopy CRCs, CRCs diagnosed before the first participant’s invitation, and those cancers diagnosed more than 24 months after screening FIT analysis were not included in the analyses.Our CRC screening program follows the Public Health laws and the Organic Law on Data Protection. All procedures performed in the study involving data from human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. No informed consent was requested since this study was based on anonymized data that is routinely collected. The study protocol was approved by the Ethics Committee for Clinical Research (PR234/16).Data regarding the monthly temperature and relative humidity (average, minimum and maximum) during the study period were registered by the Automatic Weather Stations Network of Catalonia (METEOCAT and L’Hospitalet weather station). Averages of the daily maximum temperatures and maximum relative humidity were divided into ≤24 °C and >24 °C for temperature and ≤89% and >89% for humidity.The following variables were included in the analysis: sex, age, deprivation index, date the FIT was performed, the quantitative result of the FIT, screening episode (first invitation or successive), colonoscopy findings, date of CRC diagnosis, and weather parameters (temperature, humidity, and quarter of the year). The deprivation index was calculated for the primary healthcare areas of Catalonia [28] with a scale from 0 (least deprived) to 100 (most deprived). This index uses aggregated income, health, disability, employment, and education indicators. Our CRC screening program database provided the following data: the FIT result, the number of individual participations, the date of the FIT, and colonoscopy findings. Then, this data was linked with information on procedures and diagnoses of public hospitals of Catalonia [29] (the minimum basic set of hospital discharge data (CMBD-AH). An exhaustive chart review of all the CRC cases was performed to avoid misclassifying a polyp into the CRC category. Subjects were followed until February 2017 to identify whether they were diagnosed after a FIT with a CRC (International Classification of Diseases 10th Revision (ICD-10): C18, C19, and C20: colon or rectum). Anal and appendix cancers were excluded.We examined the positivity rate, neoplasia detection rate, and interval cancer rate related to temperature and humidity. As the hemoglobin concentration was not normally distributed and logarithmic transformation did not achieve data normalization, bootstrapping techniques to calculate the mean and confidence intervals were used. A non-parametric test (Kruskal–Wallis rank test for equality) was used to evaluate the difference across quarters of the year and temperature levels. Multiple logistic regression was used to assess the influence of FIT positivity according to weather parameters, adjusting for sex, age, screening episode (first or successive screening), and deprivation index. Multiple logistic regression was also used to evaluate the probability of an interval CRC versus a screening-detected CRC according to temperature and to adjust for sex, age at diagnostic, screening episode, and deprivation index. We performed a sensitivity analysis including temperature and humidity as continuous variables, obtaining similar results. We presented the results as adjusted odds ratios (ORs) and 95% confidence intervals (CI) using logistic regression. Statistical analysis was carried out using R statistical software (R Foundation for Statistical Computing, Vienna, Austria).A total of 53,860 participants of the CRC screening program had at least one conclusive FIT result (positive or negative) (n = 92,273). Figure 1 shows a flowchart of the population examined, from when they were invited to perform the FIT to when they were diagnosed with a CRC. The FIT resulted in 5048 (5.5%) positive tests (4991 individuals), but in 530 (10.5%) of them, the colonoscopy was not performed, either because of refusal (n = 353) or it was not indicated for medical reasons (n = 177). These subjects were excluded from the analysis.Baseline characteristics of participants according to the FIT result are shown in Table 1. Screen-detected CRCs, high-risk lesions, and intermediate-risk lesions as most advanced findings were found in 211 (4.6%), 822 (18.2%), and 1118 (24.7%) participants, respectively. In 762 (16.9%) and 1605 (35.5%) participants, low-risk lesions and no preneoplastic lesions were found, respectively.Table S2 shows the main characteristics of the participants according to temperature and humidity. The median average temperature throughout the year was 14 °C, and relative humidity was 65%. The average high ambient temperature was 19 °C (25th and 75th percentiles 17, 21) in January–March; 27 °C (24, 29) in April–June; 31 °C (30, 33) in July–September; and 21 °C (18, 20) in October–December. The average high ambient humidity was 82% (25th and 75th percentiles 78, 85) in January–March; 84% (83, 85) in April–June; 84% (82, 86) in July–September; and 86% (81, 89) in October–December.A high temperature was defined as >24 °C and high humidity as >89%. Figure 2a,b shows the monthly fluctuations in positivity for the FIT over the year and the monthly temperature and relative humidity during the study period, respectively. The average temperature was highest in August, July, June, and September. Relative humidity remained remarkably stable during the year. The lowest positivity rates were in June, July, and August, which are the months with the higher temperatures. However, September had a similar temperature, and the FIT positivity was higher. The positivity rate of the FIT with a cut-off of ≥20 μg Hb/g feces was significantly lower in temperatures > 24 °C (5.2%) than in ≤24 °C (5.6%) (p = 0.005).Table 2 shows the mean hemoglobin concentration according to the seasons. The lowest concentration was recorded in the third quarter (July–September) (mean 27.2 ng/mL (95% CI 24.6 to 29.7)) and the highest in the first quarter (January–March) (mean 30.5 ng/mL (95% CI 29.1 to 32.0)).The results of the multivariate analysis (Table 3) for the positivity rate showed that the covariates associated with a higher positivity rate were male sex, advanced age, a lower socioeconomic status, and temperature > 24 °C (OR adjusted by sex, age, deprivation index, and successive screening: 0.88; 95% CI, 0.83–0.94). The multivariable analysis did not include the interaction between high ambient temperature and humidity (p = 0.83). Ambient humidity did not affect the positivity rate of FIT in the univariate or multivariable analysis. The results of the logistic regression of the probability of positive screening tests by quarter of the year (Table S3) show that there was a 15% lower probability of the FIT being positive in July–September than in January–March.We also assessed the impact of ambient temperature on the two-year FIT sensitivity. The detection rate of the FIT for advanced polyps (including screen-detected CRCs, high-risk lesions, and intermediate-risk lesions) decreased but not significantly with temperature (≤24 °C detection rate of 23.65 (95% CI 22.50–24.87) vs. >24 °C detection rate of 22.54 (95% CI: 20.85–24.35). No differences were found for humidity (≤89% detection rate of 23.25 (95% CI 22.23–24.31) vs. >89% detection rate of 23.77 (95% CI 21.03–26.86)). Finally, we observed that the probability of detecting advanced neoplasia was similar by temperature (Table 4) or different quarters of the year (Table S4).As Figure 1 shows, there were 51 interval CRCs and 211 screen-detected CRCs. We excluded the following from the analyses: 19 CRCs after a FIT positive result followed by a colonoscopy without CRC (post-colonoscopy CRC), 10 CRCs diagnosed in FIT positive participants who did not undergo a colonoscopy; and 64 diagnosed more than 24 months after screening FIT analysis (individuals with a negative FIT result who did not participate in successive screening). FIT sensitivity for CRC during this period using the interval cancer proportion method was 19.5%. Interval cancers were detected in ≤12 months of the FIT being performed in 33% of the cases (n = 17), and 67% (n = 34) of them were diagnosed after 12 months. Compared with the screen-detected group, there were no differences in the quarter of the year when the FIT was performed and the detection of interval CRCs (p = 0.11). Moreover, no differences were found when comparing screen-detected and interval CRCs according to ambient temperature (OR adjusted by sex, age, deprivation index, and successive screening: 1.72; 95% CI, 0.83–3.57).In this study, we analyzed the association between ambient temperatures and the performance of FIT in a CRC screening population. As already mentioned in the introduction, much of the previous research has been restricted to assessing the effect of ambient temperature on the FIT positivity rate. The findings of this investigation complement those of earlier studies as we have also analyzed the impact of ambient temperature and humidity on the two-year sensitivity of the FIT. The findings showed that the positivity rate of the FIT with a cut-off of ≥20 μg Hb/g feces was slightly lower when the ambient temperature was >24 °C, which is consistent with some previous reports [11,12,13,14,15]. However, monthly variations in temperature or humidity when the FIT was performed did not modify the detection rate for advanced neoplasia (CRCs, high-risk lesions, and intermediate-risk lesions). Additionally, the interval cancer detection rate was similar regardless of temperature or humidity. It is important to highlight that our screening program did not send invitations during the warmer months (summer holidays), so the July and August participation rate was very low. Moreover, we instructed participants to return the sample to the pharmacy as soon as possible and to refrigerate it. This might influence how the impact of the decrease in positivity is not causing a decrease in the detection rates.Studies focusing on the impact of ambient temperature on FIT and advanced neoplasia detection are scarce. An experimental study [30] that compared the detection rate of advanced neoplasia between low (<10 °C) and high (≥25.0 °C) temperature groups among FIT participants concluded that high ambient temperature was not a risk factor for either a positive FIT result or the detection of advanced neoplasia. In population-based screening programs, only four studies have analyzed the impact of temperature on advanced neoplasia detection rates [13,14,15,21] and two have analyzed the impact on interval cancer rates [13,21]. The study of Cha et al. [13], where subjects were instructed to return the FIT sample rapidly and store it in a refrigerator, analyzed around five million FITs of different brands and several cut-off points. Moreover, they only analyzed the data according to seasons but not by temperature. They reported that cancer detection rates were not influenced by season. When quantitative FITs were analyzed, interval cancers were more frequently detected in the summer and autumn seasons than in the winter. Nevertheless, the impact on interval cancer rates is difficult to interpret as the data included different cut-offs and FIT brands. Doubeni et al. [14] reported that the FIT’s sensitivity for CRC was significantly lower in June/July (75%) than in December/January (79%), but participants were not instructed to refrigerate the collected samples. In the Doubeni et al. [14] and the Cha et al. [13] studies, the sensitivity of FIT could be overestimated as they defined CRC diagnosis as within 12 months of the test date and not 24 months. In Niedermaier et al. [15], participants were asked to send the samples by regular mail; they examined sensitivity and specificity for advanced neoplasia at five different cut-offs and according to the maximum temperature while returning the FIT and according to the interval time to return the sample. Hemoglobin concentrations were lower in individuals with advanced neoplasia when FIT samples were exposed to ≥25 °C, compared to <10 °C. Although they observed that the differences in sensitivity stratified by the temperature at FIT higher positivity thresholds were apparently stronger, they were not significant. Finally, Bretagne et al. [21] suggested that the spring/summer was a risk factor for interval cancers but warned that this result must be interpreted cautiously as the confidence intervals were very wide (n = 209 interval cancers), and the data were from two rounds of guaiac fecal occult blood test and one round of FIT (cut-off of ≥30 μg of hemoglobin per gram of feces). Therefore, comparability with previous studies is difficult due to differences in the temperature and storage time the samples are subjected to and the country’s weather conditions.Ambient humidity did not affect FIT’s positivity rate in our screening population. This contrasts with the results of a cohort in Korea [17]. They found that high temperature and humidity were associated with a low positivity rate of FIT, but neither high temperature alone nor high humidity alone affected the positivity rate. Since we did not detect differences in positivity according to humidity in the univariate analysis, we did not include the interaction between humidity and temperature in the multivariate analysis. No noticeable differences in humidity in our study could be explained by the fact that the humidity in Spain is high but stable (around 65%) during the year, while in Korea it varies from 60% to 90% depending on the month of the year. Park et al. [17] could not perform a specific analysis with the cancer detection rate as they only included 567 subjects with advanced adenoma and 27 with CRC.Studies of FIT stability [6,7,8,9,10] have showed that exposure to ambient temperature decreased the FIT’s test performance. Some previous studies in population-based screening programs suggest, though with caution, that interval cancers rates may increase with higher temperatures. We have observed that the decrease in positivity did not impact either the detection rate of the FIT for advanced polyps or the interval cancer detection rate. Although we cannot measure the actual effect of the cold chain strategy in our findings, it can be assumed that maintaining the cold chain and reducing the period between collection and measurement during FIT performance minimized the impact of temperature on FIT sensitivity. For this reason, one of the issues that emerge from our and previous findings is the need to implement effective measures to overcome the suboptimal performance of FIT during high ambient temperatures. This is of paramount importance given that it is predicted that temperature means and extremes are projected to be higher (1.5–2 °C) in the coming years [31]. Screening programs, especially in those where the FITs are returned by mail, should implement circuits so that temperature does not impact the detection of advanced neoplasia. More effective measures could be used to optimize FIT performance, as Grazzini et al. [11] proposed, such as (a) decreasing the cut-off levels of FIT during warmer months; (b) reducing the interval of 2 years between screening episodes for those tested during warmer months; and (c) modifying the period of invitation so that a participant with a FIT performed in a warm month for the first personal screening round would be invited during the cold months for the next screening round. Of all the measures that could be adopted, we believe that the most feasible to implement would be to instruct subjects to guarantee the cold chain after sample collection, decrease the number of FITs performed during warm months and decrease the cut-off during warm periods. Other measures could increase program complexity.The generalizability of these results is subject to certain limitations. One weakness of this study that could have affected the temperature measurements is that we only had monthly temperature data and not daily. We did not have data on the time of fecal sampling to adjust for sample delay. Another concern is that temperature fluctuations in the first to the second stage of the process (sample storage at home at 2–8 °C and delivery to the pharmacy at ambient temperature) could also impact the screening results. Still, we could not investigate this pre-analytical factor in a population-based screening program. Experimental studies should evaluate this effect. On the other hand, we did not have detailed information on humidity (air conditioning or dehumidifier devices) which could also impact the results. Program invitations are distributed according to geographical areas, but these differences have been minimized by adjusting for an aggregated deprivation index. Finally, the number of CRCs cases was relatively small, although we included a large number of subjects, and the proportion of interval cancer was consistent with previous literature [32].The positivity rate of the FIT decreased with high ambient temperature. Still, it did not affect the detection rate of the FIT for advanced polyps or the interval cancer rate, probably because we have adopted measures to minimize the impact of ambient temperature on positivity rates. Humidity was not associated with a lower positivity rate. However, further prospective studies are needed to evaluate the effect of temperature specifically on the interval cancer detection rates among biennale FIT-based programs with different screening procedures.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051153/s1, Table S1: Temperature and storage time of the fecal immunochemical tests. Table S2: Baseline characteristics of participants according to temperature and humidity, respectively. Table S3: Logistic regression of positive screening tests by quarter year (95% CI). Table S4: Logistic regression of having advanced neoplasia diagnosed (screen-detected CRCs, high-risk lesions, and intermediate-risk lesions) by quarter year.Conceptualization, M.G. and G.I.-S.; methodology, M.G. and G.I.-S.; formal analysis, N.M., N.V. and G.I.-S.; investigation, G.I.-S. and M.G.; data curation, J.R., C.A., N.M., N.V. and G.I.-S.; writing—original draft preparation, G.I.-S.; writing—review and editing, M.G. and N.V.; supervision, G.I.-S., N.M., N.V., C.V., G.B., J.R., C.A., V.M., R.S.-P. and M.G.; funding acquisition, M.G. All authors have read and agreed to the published version of the manuscript.This study was funded by the Instituto de Salud Carlos III through project PI16/00588 (co-funded by the European Regional Development Fund (ERDF), a way to build Europe) and Secretaria d’Universitats i Recerca del Departament d’Empresa i Coneixement de la Generalitat de Catalunya, grant number 2017 SGR 1283. We thank the CERCA Programme/Generalitat de Catalunya for institutional support.Not applicable.Not applicable.The data presented in this study are available on request from the corresponding author.We would like to thank the Catalan Meteorological Service (Meteocat) and l’Ajuntament of L’Hospitalet de Llobregat for their administrative support. We gratefully thank the following membership of the characterization and molecular subtyping of interval cancers in colorectal cancer screening research group (MSIC-SC) in Spain: Susana López, Biobank HUB-ICO-IDIBELL, L’Hospitalet de Llobregat, Barcelona; Elvira Torné, Catalonian Health Service (CatSalut), Barcelona, Spain; Francisco Rodríguez-Moranta, Department of Gastroenterology, Bellvitge University Hospital, Hospitalet, Spain; Xavier Sanjuan, Mar Varela, Department of Pathology, Bellvitge University Hospital (HUB-IDIBELL), L’Hospitalet de Llobregat, Barcelona, Spain; Mar Iglesias, Department of Pathology, Parc de Salut Mar, Barcelona, Spain; Míriam Cuatrecasas; Department of Pathology, Hospital Clínic, Barcelona, Spain; Endoscopic Unit, Gastroenterology Department, Institut de Malalties Digestives Metabòliques, Hospital Clínic, Barcelona, Spain; Fiorella Ruiz-Pace, Oncology Data Science Group, Vall d’Hebron Institute of Oncology (VHIO), Barcelona, Spain; Francesc Macià, Prevention and Cancer Registry Unit, Service of Epidemiology and Evaluation, Parc de Salut Mar, Barcelona, Spain; Llúcia Benito, School of Nursing, University of Barcelona, Fundamental Care and Medical-Surgical Nursing Department, Hospitalet de Llobregat, Barcelona, Spain.The authors declare no conflict of interest.Flow diagram of the participants included in this analysis. CRC: colorectal cancer; FIT: Fecal immunochemical test. a Negative colonoscopy was defined as having low-risk lesions or no preneoplastic lesions.Monthly fluctuations in positivity for the FIT over the year (red line) and the monthly temperature (a) and humidity (b) during the study period, respectively. The extremes in the confidence intervals represent the minimum and maximum temperature of each month.Baseline characteristics of participants according to the FIT result.FIT: fecal immunochemical test.Variation in hemoglobin (ng/mL) concentration according to the quarter of the year when the FIT was performed.Hb: Hemoglobin (ng/mL); FIT: fecal immunochemical test; p25, p50, p75 are 25th, 50th and 75th percentiles, respectively. Kruskal–Wallis: p < 0.001.Logistic regression of the probability of positive screening tests by temperature and humidity (95% CI).1 Continuous variable; 2 All variables shown are included in the multivariate analysis. FIT: fecal immunochemical test.Logistic regression of advanced neoplasia (screen-detected CRCs, high-risk and intermediate-risk lesions).1 Continuous variable; 2 All variables shown are included in the multivariate analysis. FIT: fecal immunochemical testPublisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01154.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Liver cancer is the fourth-leading cause of cancer-related mortality worldwide and lacks effective therapies. Hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA) are the two most common types of liver cancer and both are associated with underlying inflammatory diseases. Thereby, interleukin-6 (IL-6)-mediated STAT3 signaling is critically involved in early carcinogenesis and disease progression. Here, we assessed the interplay between STAT1 and STAT3 in IL-6 signaling in vitro and studied the activation of STAT1 and STAT3 in a cohort of 124 HCC and a cohort of 138 CCA patients by immunohistochemistry. We found that IL-6 induced STAT1 transcriptional activity upon STAT3 depletion, suggesting that HCC tumor cells may activate both STAT1 and STAT3 signaling under pro-inflammatory conditions. Furthermore, HCC patient tissues showed a strong positive correlation of STAT1 and STAT3 activation in distinct patient groups. These patients also exhibited a high degree of immune cell infiltration, suggesting that these tumors are immune “hot”.Liver cancers, which are mostly hepatocellular carcinoma (HCC) and cholangiocarcinoma (CCA), are very aggressive tumors with poor prognosis. Therapeutic options with curative intent are largely limited to surgery and available systemic therapies show limited benefit. Signal transducer and activator of transcription 1 (STAT1) and 3 (STAT3) are key transcription factors activated by pro-inflammatory cytokines such as interferon-γ (IFN-γ) and interleukin-6 (IL-6). In this study, we combined in vitro cell culture experiments and immunohistochemical analyses of human HCC (N = 124) and CCA (N = 138) specimens. We observed that in the absence of STAT3, IL-6 induced the activation of STAT1 and its target genes suggesting that IL-6 derived from the tumor microenvironment may activate both STAT1 and STAT3 target genes in HCC tumor cells. In addition, STAT1 and STAT3 were highly activated in a subset of HCC, which exhibited a high degree of infiltrating CD8- and FOXP3-positive immune cells and PD-L1 expression. Our results demonstrate that STAT1 and STAT3 are expressed and activated in HCC and tumor infiltrating immune cells. In addition, HCC cases with high STAT1 and STAT3 expression also exhibited a high degree of immune cell infiltration, suggesting increased immunological tolerance.Despite growing efforts over the last two decades, liver cancer is the fourth-leading cause of cancer-related mortality worldwide [1,2]. Hepatocellular carcinoma (HCC) is the most common type of liver cancer, followed by cholangiocarcinoma (CCA), which accounts for 10–15% of liver cancers [2,3]. Both HCC and CCA can be caused by chronic hepatobiliary diseases such as chronic infection with hepatitis B (HBV) and C viruses (HCV) or other inflammatory liver diseases such as alcoholic and non-alcoholic steatohepatitis and primary sclerosing cholangitis [4], thus making HCC and CCA a paradigm for inflammation-induced carcinogenesis [5,6,7]. Thereby, tumor cells are growing in a complex microenvironment of tumorous and non-tumorous cells, and secreted small molecules [5,6]. However, the interplay between tumor cells and the tumor microenvironment including infiltrating immune cells is still poorly understood.Tumor cells are highly effective in escaping from immune-mediated eradication and they may even induce tumor promoting factors in the tumor microenvironment [8]. This interplay of tumor cells with components of their microenvironment may induce pro-tumorigenic pathways, enhancing tumor progression and leading to poor outcome [9]. In HCC and other cancer types, several studies have shown the critical involvement of inflammation, especially of interleukin-6 (IL-6) signaling during carcinogenesis [10,11,12]. In an HCC mouse model of diethylnitrosamine (DEN)-induced tumorigenesis, ablation of IL-6 expression led to lower HCC incidence in male mice, whereas female mice did not show this effect due to estrogen-mediated inhibition of IL-6 secretion [13]. Supporting these data, high serum IL-6 levels were found to be associated with rapid progression from chronic viral hepatitis to HCC in HBV- and HCV-positive patients [14,15]. Therefore, it is crucial to decipher the role of IL-6 signaling in the interplay between tumor cells and the tumor-microenvironment for the development of novel treatment modalities.Signal transducer and activator of transcription 3 (STAT3) is the main transcription factor mediating IL-6-induced signaling in tumor and immune cells, whereas interferon-γ (IFN-γ) signaling activates the closely related STAT1 protein [16]. A crosstalk between STAT1 and STAT3 has been proposed with STAT1 and STAT3 playing opposite roles in tumorigenesis [17,18]. They both modulate tumor angiogenesis, invasion, and anti-tumor immune response in an opposing manner. STAT1 is considered to be a tumor suppressor while STAT3 is rather seen as an oncogene [19,20,21]. Nevertheless, hepatoprotective functions of STAT3 during liver damage have also been reported [22]. In the tumor and its microenvironment, STAT1 and STAT3 have both been shown to be expressed by tumor cells and infiltrating immune cells and to be involved in regulating cancer adaptive immunity [21]. Recently, we demonstrated that immune cell infiltration is associated with downregulation of the tumor suppressor SH2 Domain Containing 4A (SH2D4A), which was able to suppress IL-6/STAT3 signaling [23,24]. However, the interplay between STAT1 and STAT3 signaling in liver cancer cells and the immune cell composition regarding STAT1 and STAT3 expression in the tumor cells is still unclear. Therefore, we analyzed the loss of STAT1 and STAT3 in HCC cell lines and the resulting downstream signaling effect. Furthermore, in this study, we aimed at dissecting the activation of STAT1 and STAT3 in tumor cells and the immune cell infiltrate in human HCC tissue samples.Tissue microarrays (TMA) including 124 HCC and 138 CCA tumor tissue samples were used. HCC tumor tissues of 124 patients who were surgically resected between 2006 and 2011 at the University Hospital of Heidelberg and histologically classified according to established criteria by two experienced pathologists independently (Table S1). Furthermore, CCA tissue samples were obtained from patients undergoing surgery at Heidelberg University Hospital between 1995 and 2010. In total, the CCA cohort consisted of 138 patients: 61 with intrahepatic CCA (iCCA), 45 with perihilar CCA (pCCA), and 32 with distal CCA (dCCA; Table S2). Both cohorts have been used previously [24,25,26,27,28]. Inclusion of tumor tissue for this study was approved by the institutional ethics committee (S-206/2005 and S-519/2019). The study was supported by the tissue bank of the National Center for Tumor Diseases (NCT, Heidelberg, Germany).For generation of tissue microarrays, 3 μm sections were cut and stained with hematoxylin and eosin (H&E). Representative areas from the tumor center and non-neoplastic bile duct tissue of the respective region were marked by two experienced pathologists. Tumor tissue cores with 0.6 mm diameter for the HCC or with 1.5 mm diameter for the CCA TMAs were consecutively punched out of the sample tissue blocks and embedded into a new paraffin array block using a tissue microarrayer (Beecher Instruments, Woodland, CA, USA).For the detection of STAT1, a monoclonal mouse IgG antibody directed against STAT1 (clone number C-136; dilution 1:200; Santa Cruz Biotechnology, Dallas, TX, USA) and for the detection of STAT3, a monoclonal rabbit IgG antibody directed against STAT3 (clone number 79D7; dilution 1:400; Cell Signaling Technology, Danvers, MA, USA), were used. Staining was performed on an automated system (Discovery Ultra, Ventana, Tuscon, AZ, USA) following the manufacturer’s instructions using Dako Target Retrieval Solution, Citrate pH 6 (S2369, Agilent Technologies, Santa Clara, CA, USA) and the Dako Target Retrieval Solution, pH 9 (S2367, Agilent Technologies), respectively. For staining of PD-L1, anti-PD-L1 (ready-to-use, clone SP263, Roche Diagnostics, Rotkreuz, Switzerland) antibody was used. Three µm sections of the TMA were deparaffinized, pre-treated with an antigen retrieval buffer (Tris/Borat/EDTA, pH 8.4; Ventana, Roche), and stained using an automated device (Ventana Benchmark Ultra, Roche). Immunohistochemistry to detect and count specific inflammatory cells using antibodies specific to CD3, CD4, CD8, CD20, CD68, CD117, and FOXP3 was performed as recently described [24,25].STAT1 and STAT3 expression were scored by applying a semi-quantitative immunoreactive score (IRS), resulting in the expression of values ranging from 0 (no expression) to 12 (strong expression in more than 80% of tumor cells), as previously described [29]. Briefly, staining intensity (0: no staining, 1: weak staining, 2: moderate staining, 3: strong staining) as well as percentage of positively stained cells (0: no cells stained, 1: up to 10% of cells stained, 2: 10–50% of cells stained, 3: 51–80% of cells stained, 4: more than 80% of cells stained) were scored separately and the IRS for each individual case was calculated by multiplication of the intensity and the percentage score. PD-L1 expression was categorized by tumor proportion score (TPS), which was defined as the percentage of tumor cells with membranous PD-L1 staining based on all tumor cells, and combined positive score (CPS), defined as the number of PD-L1-positive cells (tumor cells, lymphocytes, and macrophages) divided by total number of tumor cells × 100%. Patients were assigned to the high expression group with PD-L1 TPS or CPS larger or equal 1%. Immunohistochemical staining of immune cell markers was performed and analyzed previously, as described. [24,25,30]. These data on immune cell infiltration were thus derived from previous studies. For technical reasons, some tissue microarray dots could not be evaluated due to loss of tissue or staining artifacts. Therefore, the number of included cases for the immunohistochemical analysis of STAT1 and STAT3 expression and for the immune cell types varied slightly. Cell lines were obtained from ATCC (HepG2) or JCRB (HuH1 and HuH7), regularly tested for mycoplasma contamination (MycoAlert, Lonza, Basel, Switzerland), and authenticated by STR analysis. HuH1 and HuH7 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) and HepG2 in RPMI-1640 medium. All growth media were supplemented with 10% FCS and 1% penicillin/streptomycin (Thermo Fisher Scientific, Waltham, MA, USA). For siRNA-mediated knockdown of STAT1 or STAT3, Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific) was used according to the manufacturer’s instructions. siRNAs were obtained from Qiagen and are listed in Supplementary Table S3.Total RNA was extracted from liver cancer cell lines by applying the NucleoSpin RNA Kit (Macherey-Nagel, Düren, Germany) according to the manufacturer’s protocols. cDNA was synthesized from 500 ng total RNA using the RevertAid H Minus First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). Samples were analyzed in triplicate using primaQUANT CYBR-Green Mastermix low ROX (Steinbrenner Laborsysteme, Wiesenbach, Germany) on a StepOnePlus real-time PCR instrument (Applied Biosystems, Darmstadt, Germany). The reference gene Serine and Arginine rich Splicing Factor 4 (SRSF4) was used as the normalization control. Relative mRNA expression values were calculated using the comparative Ct method. Primers were obtained from Thermo Fisher Scientific and Apara-Bioscience (Denzlingen, Germany) and are listed in Supplementary Table S4.Total protein was extracted from cells using cell lysis buffer 10× (Cell Signaling Technology) supplemented with phosphatase inhibitor PhosStop and protease inhibitor Complete Mini EDTA-free (both Roche Diagnostics, Mannheim, Germany). Protein concentrations were determined using the Bradford assay (Sigma Aldrich, Taufkirchen, Germany). Protein samples were prepared in equal amounts with water and 4× loading buffer (250 mM Tris pH 6.8, 8% SDS, 40% glycerol, 100 mM DTT, 0.04% bromophenol blue). Twenty µg of protein was separated on 8% to 12% Bis/Tris-polyacrylamide gels and then transferred to an equilibrated nitrocellulose membrane (Merck Chemicals, Darmstadt, Germany). Membranes were blocked with 5% milk in Tris-buffered saline with Tween 20 (TBST) or 5% bovine serum albumin (BSA) in TBST and incubated with the indicated primary antibodies (Supplementary Table S5) overnight at 4 °C. Proteins were detected with IRDye secondary antibodies using an Odyssey Sa Infrared Imaging System (LI-COR Biosciences, Bad Homburg, Germany). Protein bands were quantified by densitometry using the Image Studio Software v.3.1.4 (LI-COR) and normalized to loading control β-Actin or β-Tubulin, as indicated. Whole Western blot images are provided in the Supplemental online.To analyze the impact of IL-6 on STAT1 transcriptional activity in STAT3-depleted cells, luciferase reporter assays were performed. HepG2 and HuH7 cells were seeded in 24-well plates and transfected with respected siRNAs by using Lipofectamine RNAiMAX Transfection Reagent. Twenty-four hours after siRNA transfection, cells were co-transfected with Firefly luciferase reporter vector pGL4 [luc2P/GAS-RE/Hygro] and pRL-TK (Renilla luciferase control reporter vector). The next day, cells were treated for 24 h with 20 ng/mL IL-6 and luciferase activity was analyzed by the Dual-Luciferase Reporter Assay System (Promega, Mannheim, Germany) according to the manufacturer’s protocol using an Omega FLUOstar Microplate Reader (BMG LABTECH GmbH, Ortenberg, Germany). Renilla luciferase was used as the internal transfection control and for normalization.In the case of two group comparisons, differences were assessed using the nonparametric Mann–Whitney U test. The association of two variables was assessed using nonparametric Spearman’s correlation analysis. Statistical analyses were performed with GraphPad Prism 6.0 (GraphPad Software, La Jolla, CA, USA) and the statistical computing environment R version 4.0.3 (http://www.R-project.org/, released 10 October 2020, last accessed 10 January 2022). For data analysis, R package Hmisc was used and plots were generated by package Corrplot. p-values below 0.05 were considered statistically significant.To assess the interplay of STAT1 and STAT3 protein in liver cancer cells, we first investigated the effect of STAT1 protein knockdown in the three liver cancer cell lines HepG2, HuH1, and HuH7, which all express the STAT1 and STAT3 proteins (Figure 1 and Supplementary Figure S1). By using two independent siRNAs against STAT1, its mRNA expression was depleted >90% in the different liver cancer cell lines (Figure 1A). Expression of IL-6/STAT3 target genes Transthyretin (TTR) and Serine Peptidase Inhibitor Kazal Type 1 (SPINK1) showed that STAT1 depleted cells treated with IFN-γ, a known inducer of the STAT1-signaling pathway, exhibited no significant induction of JAK/STAT3-pathway activation at the transcriptional level in all three cell lines (Figure 1A). At protein level, STAT1-depleted HepG2 cells showed a slightly stronger Tyr705 phosphorylation of STAT3 (pSTAT3 Tyr705) after 2 h of IFN-γ stimulation compared to Allstars transfection control (Figure 1B, top and bottom panel). However, the activation of STAT3 by Tyr705 phosphorylation did not persist and was not observed in HuH1 (Figure S1A) and HuH7 (Figure S1B) cells, suggesting that STAT1 depletion does not induce an IL-6-like response in HCC cells. Therefore, we did not observe any significant effect of STAT1 protein depletion on STAT3 protein activation or on STAT3 target gene expression in three different liver cancer cell lines.Next, we performed the opposite experiment by depleting STAT3 protein expression followed by IL-6 treatment. In three different liver cancer cell lines, HepG2, HuH1, and HuH7, siRNA-mediated STAT3 depletion prolonged STAT1 phosphorylation at Tyr701 upon IL-6 treatment compared to Allstars transfection control (Figure 2 for HuH1 and HuH7 and Supplementary Figure S2A for HepG2). In control cells, short-term activation of STAT1 represented by phosphorylated STAT1 at Tyr701 decreased from 1 h after stimulation and stayed very low. In contrast, pSTAT1 Tyr701 was still observed at 2 h, 4 h, and 24 h time points in STAT3-depleted cells. Interestingly, after 24 h, the expression of total STAT1 was increased compared to the control cells, indicating a positive feedback mechanism (Figure 2).Next, we aimed to analyze the impact of STAT3 depletion on STAT1 transcriptional activity by applying the luciferase assay (Figure 3A). While control cells showed no transcriptional activation upon IL-6 treatment, STAT3-depleted cells showed increased activity using siSTAT3#1 (Figure 3B,C). However, siSTAT3#2 did not show any increased luciferase activity potentially due to inefficient deletion of STAT3 protein (Figure 2A,B and Figure 3D,E). To further evaluate the effect of potential STAT1 transcriptional activity upon STAT3 depletion, we evaluated the gene expression levels of well-known STAT1 target genes Interferon Regulatory Factor 1 (IRF1) and Apolipoprotein L1 (APOL1). In HuH1 and HuH7 cells, both STAT1 target genes were significantly upregulated upon IL-6 stimulation in STAT3-depleted cells (Figure 3D,E for HuH1 and HuH7; Supplementary Figure S2B for HepG2). Interestingly, STAT1 expression was also significantly induced, suggesting a positive feedback loop (Figure 3D,E). Therefore, downregulation of STAT3 protein levels results in increased transcriptional activity of STAT1 in the presence of IL-6 and prolonged activation of STAT1 signaling.Immune cell infiltration has been reported to influence epithelial tumor cells [21]. As we found that IL-6 induces STAT1 transcriptional activity upon STAT3 depletion, we next sought to analyze the expression of STAT1 and STAT3 regarding infiltrating immune cells in human liver cancer tissues. HCC (N = 124) and CCA (N = 138) tumor tissue microarrays were stained for STAT1 and STAT3 using immunohistochemistry (Tables S1 and S2). Comparison of STAT1 and STAT3 nuclear expression levels, respectively, in tumor cells revealed no significant differences in age, sex, etiology, liver disease, nodularity, vascular invasion, and grading in patients with HCC (Table S1). High nuclear expression of STAT3 protein was associated with male gender (p = 0.0002) and perihilar or distal localization (p = 0.001) in CCA but not with histology, grading, and tumor staging (Table S2). In addition, clinical characteristics of patients with CCA did not differ between STAT1 low and STAT1 high groups (Table S2).Next, T-cell populations were quantified by anti-CD3, anti-CD4, anti-CD8, and anti-FOXP3 staining. B cells, macrophages, and mast cells were detected using anti-CD20, anti-CD68, and anti-CD117 staining, respectively. To evaluate the potential association of STAT1 and STAT3 with specific infiltrating immune cell populations, we performed paired correlation analyses. Overall, strong positive correlations were observed in HCC (Figure 4), but fewer correlations were seen in CCA (Supplementary Figure S4). Thereby, total immune cell counts and intraepithelial immune cell counts exhibited similar effects in CCA (Supplementary Figure S4). In HCC, nuclear and cytoplasmic STAT1 expression, together with the STAT1-positive immune cell infiltrate, correlated highly significant with CD3-, CD4-, CD8-, and FOXP3-positive immune cell infiltrates (Spearman r = 0.335, 0.349, 0.348, and 0.394, all with p < 0.001; Figure 4A,B and Supplementary Figure S3). In addition, weaker, but significant correlations of nuclear and cytoplasmic STAT1 in the tumor cells and of STAT1-positive immune cells were observed with the CD20- and CD68-positive immune cell infiltrate (Figure 4A). Only few CD117-positive immune cells were detected and no significant correlation was observed for CD117-positive immune cell infiltrate with nuclear or cytoplasmic STAT1 expression of the tumor epithelium. In contrast, correlation coefficients for STAT3 were lower overall and reached significance in fewer pairings compared to STAT1. Thus, the highest correlation existed between epithelial nuclear STAT3 staining and CD4- or FOXP3-positive immune cells (Figure 4A).As tumors with high immune cell infiltration, also termed immune “hot” tumors, have been suggested to express the immune checkpoint molecule PD-L1 [31], we correlated the expression of PD-L1 with immune cell population markers and STAT protein expression (Figure 4). PD-L1 TPS, indicating tumor-specific, and PD-L1 CPS, indicating immune cell and tumor-specific staining, both strongly correlated with STAT1 expression and with cytotoxic CD8- and FOXP3-positive cells (Figure 4). Thus, PD-L1-positive HCC exhibited immune “hot” characteristics with significantly higher overall immune cell infiltration (Figure 5). Furthermore, when comparing HCC tissue samples by median of nuclear STAT3 expression, CD3-, CD4-, CD8-, and FOXP3-positive immune cell counts were significantly higher in tumor tissues with high nuclear STAT3 expression compared to tumor tissues with low nuclear STAT3 expression of the tumor cells (Figure S5). No difference between HCC tumor cells with high or low nuclear STAT3 tumor epithelium was detected regarding CD20-, CD68-, and CD117-positive immune cell infiltrates (Figure S5). Therefore, active STAT3 signaling in the tumor cells, evident by nuclear localization of STAT3, was associated with infiltrating FOXP3-T-cell numbers, suggesting that nuclear STAT3 may be associated with increased active immune response.To better understand the role of STAT3 in the tumor microenvironment, we compared HCC samples that were positive or negative for STAT3-expressing immune cells. Patient samples with high numbers (median cutoff) of STAT3-expressing immune cells had significantly higher infiltration of CD3-, CD4-, and CD20-positive immune cells (Figure 6A). Interestingly, STAT1 and STAT3 exhibited highly positive correlations in the tumor epithelium. Nuclear STAT1 was strongly and significantly correlated with cytoplasmic STAT1, nuclear STAT3, and cytoplasmic STAT3, and vice versa. HCC tissues with positivity for tumor cell nuclear STAT1 had high cytoplasmic STAT1, suggesting our in vitro data showed that prolonged activation of STAT1 phosphorylation results in increased total STAT1 protein levels (Figure 6B). Similarly, tumor tissue samples with high positivity for tumor cell nuclear STAT3 exhibited high cytoplasmic STAT3 levels and tumors with positivity for nuclear STAT1 showed increased nuclear STAT3 (Figure 6C,D). Therefore, activated STAT1 and activated STAT3 signaling in the tumor cells was observed in a subsets of HCC patients and may indicate a specific HCC subgroup with high immune cell infiltration.The tumor microenvironment plays a crucial role in cancer progression, and targeting the immune system by cancer immunotherapy has become a promising therapeutic option recently approved for HCC. However, in HCC and several other tumor entities, cancer immunotherapy is effective only in a small subpopulation of patients. Therefore, it is important to better understand the composition of distinct immune cell populations and their interaction with the tumor cells within the tumor microenvironment. This study focused on the role of STAT1 and STAT3 signaling in liver cancer. Using human cell lines, we demonstrated that IL-6 induces STAT1 transcriptional activity upon STAT3 depletion, suggesting that HCC cells may activate both STAT1 and STAT3 signaling under pro-inflammatory conditions. Consistently with these in vitro observations, HCC subgroups showed a great degree of positive correlation of STAT1 and STAT3 activation. These STAT1-positive tumors also exhibited a high degree of immune cell infiltration, especially CD4-, CD8, and FOXP3-positive T cells, suggesting that an active engagement of the immune system with the tumor takes place and that these tumors are immune “hot”. Our observation that STAT1-signaling and STAT1 target gene expression may be activated by IL-6 proposed that STAT3 and STAT1 signaling are interconnected in HCC. This is also supported by the positive correlation of STAT1 and STAT3 expression in human HCC samples in tumor and immune cells, indicating high immunological tolerance in a subset of HCC patients, as evident by PD-L1 expression.Interestingly, STAT3 knockout in mouse embryo fibroblasts exhibited an IFN-γ-like response including an extended STAT1 activation upon IL-6 treatment (Costa-Pereira et al., 2002). In contrast, upon STAT1 loss, IFN-γ-mediated STAT3 activation was much stronger and prolonged, leading to the expression of some STAT1 target genes that are typically transcribed in IFN-γ-treated wild-type cells [32]. Similar effects were also seen in STAT1-deficient bone-marrow-derived macrophages and T lymphocytes [33]. Thus, our data in liver cancer cell lines are in line with previous studies in fibroblasts and ma-crophages whereby a disbalance between STAT1 and STAT3 may lead to the activation of IFN-γ/STAT1 target genes in a proinflammatory IL-6-containing tumor microenvironment. To some extent, this may explain the pro- and anti-tumorigenic effects of STAT1 signaling in liver carcinogenesis [18]. Furthermore, we could show that STAT1 depletion did not induce an IL-6-like response in HCC cells treated with IFN-γ, indicating distinct regulatory mechanisms. Taken together, STAT1 and STAT3 signaling pathways are tightly interconnected and IFN-γ/STAT1 target genes may be activated in a proinflammatory IL-6-containing tumor microenvironment.Furthermore, we analyzed the expression of STAT1 and STAT3 protein in human HCC and CCA tissue samples separately. STAT1 and STAT3 protein expression were strongly and positively correlated in the tumor cells and infiltrating immune cells. This suggests that the stimulatory cytokines in the tumor microenvironment may equally act on the tumor cells and the immune cells, and that various cell types are susceptible for inflammatory signals.Tumor-infiltrating T cells are a marker of increased immunological tolerance in cancer. Tumors with the so-called T cell–inflamed phenotype consisting of infiltrating T cells appear to resist immune attack through the dominant inhibitory effects of immune system–suppressive pathways [34]. In contrast, it is believed that tumors that lack T-cell infiltration may resist immune attack through immune system exclusion or ignorance [34]. A recent large-scale profiling study aiming to identify immune subtypes across 33 cancer types suggested six immune subtypes with prognostic relevance [9]. Thereby, substantial infiltration of CD8- and CD4-positive immune cells was observed in four out of six subtypes, stressing the importance and wide involvement of CD4- and CD8-positive T cells. Furthermore, infiltrating T cells have prognostic and therapeutic relevance. Higher presence of tumor-infiltrating T lymphocytes is generally considered a favorable prognostic factor [35]. In CCA, patients with intraepithelial tumor-infiltrating CD4-, CD8-, and FOXP3-T lymphocytes showed a significantly longer overall survival [25]. Similarly, significantly fewer patients with a high density of CD8-positive T-cell infiltrates experienced recurrence of their HCC within three years compared with those exhibiting a low CD8-cell density [36]. Despite the favorable prognosis of patients with T-cell infiltration, long-term survival of patients with or without T-cell infiltration is dismal.FOXP3-positive regulatory T-cells (Treg) are believed to mediate the suppression of anti-tumor immunity, which may lead to more aggressive disease [37,38]. It has been suggested that STAT3 directly binds to a STAT consensus site in the FOXP3 promoter to enhance FOXP3 expression of Treg cells increasing their inhibitory function [39,40]. Here, we demonstrated that STAT3 expression and activation, evident by nuclear localization, strongly correlated with the presence of FOXP3-positive Tregs. The development of immune checkpoint inhibitors that block negative regulators of T-cell immunity provide powerful therapeutic options. However, immunotherapies are only effective in a small subset of liver cancer patients and a better understanding of the tumor microenvironment is crucial to improve patient outcome. Interestingly, STAT3 may directly bind the PD-1 promoter and activate PD-1 protein expression in T cells [41]. Therefore, the degree of STAT3-positive immune cells in HCC may be linked to the response to immune checkpoint blockade.Our observation that STAT1 and STAT3 expression exhibit strong positive correlation in the tumor cells and the immune cell infiltrate suggest that both may similarly indicate the state of the tumor microenvironment. Existing evidence suggests that JAK-STAT signaling members may ultimately serve as diagnostic markers stratifying patients that may benefit from more targeted therapeutic approaches that modulate downstream targets, rather than upstream JAK-STAT pathway regulators [42]. Targeting oncogenic transcription factors of the STAT family has been suggested as powerful approach potentially modulating gene regulatory processes including chromatin remodeling [43]. In contrast to other tumor entities, PD-L1 blockage did not significantly prolong HCC patient survival, suggesting that PD-L1/PD-1 axis blockade alone may not be sufficient to initiate adequate levels of anticancer immunity in HCC [44,45]. Recently, it has been demonstrated that inhibition of STAT3 leads to a reduction in PD-L1 protein expression, which may constrain tumoral inflammation and improve immune response against tumor cells [46,47]. Furthermore, we were able to show that PD-L1 and STAT1 expression correlated in both the tumor cells and the tumor infiltrating immune cells. Interestingly, IL-6 knockdown in cancer associated fibroblasts (CAF) increased IFN-γ on CD8-postive T cells and IL-6 blockade could reverse anti-PD-L1 resistance in an HCC mouse model [48]. Recently, a comprehensive study in non-small cell lung cancer demonstrated that reduction in STAT3 in the tumor microenvironment using an antisense oligonucleotide reversed immunotherapy resistance in preclinical STK11 knockout mouse models [49]. Supporting these data, inhibition of the AURKA/STAT3 signaling pathway promoted effective T-cell infiltration into the tumor microenvironment and improved anti-PD-1 efficacy [50]. Therefore, a combination of STAT3-inhibition and immune therapy may be beneficial to HCC patients [31,45]. However, clinical studies are not available thus far and the role of STAT1 and STAT3 is not yet fully understood [51]. Thus, it will be crucial to evaluate different approaches to inhibit STAT3 signaling in the tumor cells and in the tumor microenvironment alone or in combination with immune therapy.In conclusion, we show that in the absence of STAT3, IL-6 induced prolonged STAT1 signaling and expression of STAT1 target genes, which suggests an interplay of STAT1 and STAT3 signaling in the presence of proinflammatory IL-6 in the tumor microenvironment. Furthermore, activation of STAT1 and STAT3 in the tumor cells strongly correlates with the activation of STAT1 and STAT3 in infiltrating immune cells and infiltration of CD4-, CD8-, and FOXP3-positive immune cells, indicating high immunological tolerance in a subset of HCC patients, as evident by PD-L1 expression. Therefore, approaches to inhibit STAT3 signaling alone or in combination with immune therapy may improve patient outcome.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051154/s1, Figure S1: STAT1 depletion does not induce IL-6-like response in HCC cells; Figure S2: STAT3 depletion prolongs IL 6 induced STAT1 phosphorylation in HepG2; Figure S3: Pairwise correlation of nuclear STAT1 staining with infiltrating immune cells in human HCC tissues; Figure S4: Correlation of STAT1 and STAT3 expression with immune cell infiltration in CCA; Figure S5: Association between nuclear STAT3 expression in the tumor epithelium and immune cell infiltration in HCC tumor samples; Table S1: Clinicopathological characteristics of the HCC cohort (N = 124) and comparison of patients with STAT1 or STAT3 nuclear low or high staining; Table S2: Clinicopathological characteristics of the CCA cohort (N = 138) and comparison of patients with STAT1 or STAT3 nuclear low or high staining; Table S3: siRNAs used for gene silencing; Table S4: Primers used for qRT-PCR; Table S5: Antibodies used for Western blot (WB) and immunohistochemistry (IHC); Supplemental Whole Western Blot Figures.Conceptualization, C.P., J.S. and S.R.; Methodology, C.P., J.S., T.H. and A.F.; Formal analysis, C.P., J.S., T.H., A.C., B.G. and S.R.; Resources, T.A., J.J., S.S., K.B., S.P., B.C.K., C.S., P.S., A.M. and B.G.; Writing—original draft preparation, C.P., J.S. and S.R.; Writing—review and editing, All authors; Visualization, C.P. and J.S.; Supervision, B.G. and S.R.; Project administration, S.R. All authors have read and agreed to the published version of the manuscript.This work was supported by Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—Project-ID 314905040-SFB/TRR 209 Liver Cancer (B01 to SR, B09 to SS and Z01, INF to PS) and the European Union’s Horizon 2020 research and innovation program under Eurostars (grant E! 113707, LiverQR) to PS and SR. SR was supported by funds from German Cancer Aid (Deutsche Krebshilfe, project no. 70113922).The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of the Medical Faculty of Heidelberg University (S-206/2005, 19 April 2013 and S-519/2019, 2 August 2019).Written and verbal consent was provided by all participants.Additional datasets analyzed during the current study are available from the corresponding author upon reasonable request.The authors thank Veronika Geissler (Institute of Pathology, Heidelberg University) for the excellent technical assistance. Samples were provided by the tissue bank of the National Center for Tumor Diseases (NCT, Heidelberg, Germany) in accordance with the regulations of the tissue bank and the approval of the Ethics Committee of the Medical Faculty of Heidelberg University.P.S. received funding for grants, boards, and presentations from Novartis. The other authors declare no competing interests. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.STAT1 depletion does not induce an IL-6-like response in HCC cells. (A) STAT1 was depleted using two different siRNAs, siSTAT1#1 and siSTAT1#2, in HepG2, HuH1, and HuH7 cells (as indicated) and cells were incubated with IFN-γ (500 U/mL, 24 h). Relative RNA expression values are shown for STAT1, STAT3, TTR, and SPINK1. Untreated control (UTC), Allstars control (AS), n = 3, * p < 0.05, ** p < 0.01, *** p < 0.001. (B) Western blot detecting STAT1, STAT3, and pSTAT3 Tyr705 in the control or STAT1-depleted HepG2 cells upon incubation with IFN-γ (500 U/mL) for indicated time points. β-Actin served as the loading control.STAT3 depletion prolongs IL-6-induced STAT1 phosphorylation. (A,B) Western blot of STAT3, pSTAT1 Tyr701, and STAT1 proteins are shown. STAT3 was depleted using two different siRNAs, siSTAT3 #1 and siSTAT3 #2 in (A) HuH1 and (B) HuH7 cells. Cells were incubated with IL-6 (20 ng/mL) for indicated time points. β-Tubulin served as loading control. (C,D) Relative quantification of STAT3, STAT1, and pSTAT1 Tyr701 protein expression levels at indicated time points from two independent experiments in (C) HuH1 and (D) HuH7 cells. Results of siSTAT3 #1 and siSTAT3 #2 are combined. * p < 0.05, ** p < 0.01, *** p < 0.001.IL-6 induces STAT1 transcriptional activity upon STAT3 depletion. (A) To quantitatively measure STAT1 transcriptional activity, a luciferase reporter containing four interferon-gamma activated sites (GAS) was used. (B,C) Luciferase assay in (B) HepG2 and (C) HuH7 cells measuring endogenous STAT1 transcriptional activity using the GAS-element containing reporter co-transfected with Allstars (control), siSTAT3 #1, or siSTAT3 #2, in the presence or absence of 20 ng/mL IL-6 for 24 h, as indicated. Renilla luciferase was used as the internal transfection control and for normalization. Data were normalized to IL-6 unstimulated transfection control cells (mean ± SD). (D,E) Relative mRNA expression levels of STAT3, STAT1, and the STAT1 target genes IRF1 and APOL1 measured by semi-quantitative RT-PCR in (D) HuH1 and (E) HuH7 cells upon transfection with Allstars control siRNAs (AS), siSTAT3 #1, or STAT3 #2 in the presence of 20 ng/mL IL-6 for 24 h. N = 3. * p < 0.05, ** p < 0.01, *** p < 0.001.Correlation of STAT1 and STAT3 expression with immune cell infiltration in HCC. (A) Correlation matrix of immune cell infiltrates and immunohistochemical STAT1 or STAT3 expression in HCC tumor tissue samples. Red color indicates negative correlation, whereas blue color indicates positive correlation. Darker coloration indicates a higher Spearman correlation coefficient and asterisk denotes the level of significance. Spearman p-value: * <0.05; ** <0.01; *** <0.001. (B) Exemplary HCC tumor tissues with low or high immunohistochemical reaction against STAT1 are shown, respectively. The corresponding immune cell infiltrate stained for CD3, CD4, CD8, FOXP3, and PD-L1 protein expression in the same tumor dot is shown as indicated. Scale bar: 100 µm.Association between PD-L1 expression in the tumor epithelium and immune cell infiltration in HCC tumor samples. (A) CD3-, (B) CD4-, (C) CD8-, (D) FOXP3-, (E) CD20-, (F) CD68-, (G) STAT1-, and (H) STAT3-positive immune cell infiltrate in HCC tumor tissue samples stratified by low (N = 90) and high PD-L1 TPS expression score (N = 15). TPS—tumor proportion score. * p < 0.05; ** p < 0.01; *** p < 0.001.Association between STAT1 and STAT3 expression in the tumor epithelium and immune cell compartment in HCC tumor samples. (A) CD3-, CD4-, CD8-, and CD20-positive immune cell infiltrate in HCC tumor tissue samples stratified by median count of STAT3-positive immune cells. (B) Cytoplasmic STAT1 expression IRS in HCC tumor samples stratified by median nuclear STAT1 expression. (C) Cytoplasmic STAT3 expression IRS stratified by median nuclear STAT3 expression. (D) Nuclear STAT3 expression IRS stratified in patient samples grouped by median nuclear STAT1 expression. * p < 0.05; ** p < 0.01; *** p < 0.001.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01155.txt

@@ -0,0 +1 @@
+Glioblastomas and brain metastases are the most common malignant intracerebral tumors in adults and are often difficult to differentiate using conventional MRI. Common to both is contrast enhancement on T1w post-Gd sequences and perilesional T2 hyperintensity, which in glioblastomas corresponds to diffuse tumor infiltration rather than to edema alone. Characterization of perifocal T2 hyperintensity in glioblastoma has been poor using MRI, and tumor recurrences frequently occur in these areas. Using a novel diffusion microstructure imaging (DMI) approach, we observed that in metastases, perilesional T2 hyperintensity is more likely due to edema alone with a higher fraction of free water and reduced cellular compartments compared to glioblastomas. DMI might be a powerful diagnostic tool, as it could be used to distinguish metastases from GBM, as well as characterizing perifocal T2 hyperintensity in GBM.Purpose: Glioblastomas (GBM) and brain metastases are often difficult to differentiate in conventional MRI. Diffusion microstructure imaging (DMI) is a novel MR technique that allows the approximation of the distribution of the intra-axonal compartment, the extra-axonal cellular, and the compartment of interstitial/free water within the white matter. We hypothesize that alterations in the T2 hyperintense areas surrounding contrast-enhancing tumor components may be used to differentiate GBM from metastases. Methods: DMI was performed in 19 patients with glioblastomas and 17 with metastatic lesions. DMI metrics were obtained from the T2 hyperintense areas surrounding contrast-enhancing tumor components. Resected brain tissue was assessed in six patients in each group for features of an edema pattern and tumor infiltration in the perilesional interstitium. Results: Within the perimetastatic T2 hyperintensities, we observed a significant increase in free water (p < 0.001) and a decrease in both the intra-axonal (p = 0.006) and extra-axonal compartments (p = 0.024) compared to GBM. Perilesional free water fraction was discriminative regarding the presence of GBM vs. metastasis with a ROC AUC of 0.824. Histologically, features of perilesional edema were present in all assessed metastases and absent or marginal in GBM. Conclusion: Perilesional T2 hyperintensities in brain metastases and GBM differ significantly in DMI-values. The increased free water fraction in brain metastases suits the histopathologically based hypothesis of perimetastatic vasogenic edema, whereas in glioblastomas there is additional tumor infiltration.Glioblastoma (GBM; IDH-wildtype (IDH-wt)) is the most common malignant primary intracerebral tumor in adults [1]. Despite intensive research efforts and current standard therapy with radical tumor resection and adjuvant radiochemotherapy, the prognosis is poor with a median survival of 14.6 months [2]. Tumor infiltration in GBM usually extends well beyond the contrast-enhancing and even T2 hyperintense tumor portions, and to date has been difficult to characterize by conventional MR imaging. The assumption that the contrast-enhancing margin of GBM does not correspond to its tumor boundary is supported by the fact that most tumor recurrences occur within peripheral T2/FLAIR hyperintense areas [3]. It is, therefore, necessary to further characterize infiltrative GBM tumor components by MRI.Brain metastases are the most common intracranial malignant tumors in adulthood and predominantly arise from five primary tumors [4] (lung cancer, breast cancer, malignant melanoma, gastrointestinal tract tumors, and renal cell cancer) and may show marked heterogeneity on MR imaging. In particular, cerebral metastases with marginal contrast enhancement, perifocal T2/hyperintensity, and solitary localization often cannot be reliably distinguished from high-grade gliomas such as GBM by conventional MR imaging.Since the treatment regimens and prognoses of GBM and metastases differ, a reliable non-invasive differentiation is of great relevance. In GBM, the goal is radical complete tumor excision of contrast-enhancing tumor portions, whereas in metastases, the focus is on systemic diagnosis and therapy. Therefore, in cases of uncertainty, stereotactic tumor biopsy is usually performed, which carries risks depending on factors such as tumor location and patient age and always involves the possibility of sampling error [5,6]. It would be helpful in clinical routine to be able to distinguish GBM and metastases more precisely with noninvasive techniques such as MRI.Advanced diffusion imaging has undergone steady development within the past decades from diffusion tensor imaging (DTI) to novel techniques such as neurite orientation and dispersion imaging (NODDI) [7], and diffusion microstructure imaging (DMI) [8]. Numerous studies have used diffusion imaging techniques to try to distinguish between brain metastases and GBM/high-grade gliomas [9,10,11] under the assumption that diffusion parameters differ between predominantly vasogenic edema, such as those in brain metastases, and vasogenic edema with additional tumor infiltration, such as those in GBM [12,13]. However, due to divergent results, these parameters are not established in routine diagnostics yet [9,14,15]. For example, Wang and colleagues found no significant differences in fractional anisotropy (FA) in the peritumoral edema between GBM and brain metastases, and they concluded that no DTI–metric threshold was able to distinguish between peritumoral infiltrative edema and vasogenic edema [11]. Another recent study employed various diffusion techniques including NODDI, DTI, and classical DWI and was able to identify significant differences within the contrast-enhancing tumor regions, but no significant alterations within the perilesional T2 hyperintense area [16].DMI is a novel technique, similar to NODDI. It allows the differentiation of three microstructural components based on their different diffusion properties. Here, the intra-axonal volume compartment can be distinguished and relatively quantified from the extra-axonal intracellular and extracellular free water compartments [7,8,17]. While water molecules in neuronal structures or the extracellular matrix are aligned by organelles and membranes, free water molecules are modelled to move randomly and unimpeded. From this, DMI derives a standard model for white matter consisting of the intra-axonal volume fraction (V-intra) with nearly one-dimensional molecular diffusion due to the narrow membrane boundaries, and the extra-axonal volume fraction (V-extra) corresponding to the extra-axonal cellular compartment and the extracellular matrix and characterized by restricted diffusion and the free water/CSF fraction (V-CSF). DMI has recently been used in clinical research to study normal pressure hydrocephalus (NPH) [18] and temporal lobe epilepsy [19] and is a promising diffusion imaging technique that could find application in numerous white matter pathologies.Considering the different edema characteristics in GBM and metastases (infiltrative vs. vasogenic), the aim of this study was to analyze, whether the perilesional diffusion microstructure imaging metrics V-intra, V-extra, and V-CSF differ between GBM and metastases and to correlate the findings histopathologically.Patients presenting with an intra-axial contrast-enhancing lesion were retrospectively included within a one-year period (2020–2021). The prerequisite for inclusion was the presence of a contrast-enhancing intra-axial lesion, with perilesional T2 signal elevations in the white matter. Patients with severe small vessel disease (Fazekas > 1), concomitant vascular lesions (e.g., vascular malformations), or neurodegenerative disorders (e.g., Alzheimer’s disease, frontotemporal lobar degeneration (FTLD), cerebral amyloid angiopathy) were excluded. Further exclusion criteria were previous radiation therapy, resection, or biopsy and poor image quality due to motion artifacts.Imaging was performed with a dedicated tumor protocol on a 3 Tesla scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) using a 64-channel head and neck coil including a diffusion tensor/microstructure imaging (DTI/DMI) sequence and isotropic T2w FLAIR and T1w MPRAGE sequences for anatomical delineation and segmentation. Post-contrast T1w sequences were acquired 4–5 min after intravenous injection of 0.1 mmol/kg gadobutrol (ProHance®, Bracco Imaging, Milan, Italy). The sequence parameters are presented in Table S1.The study was conducted in accordance with the 1964 Helsinki Declaration and its later amendments and approved by the local ethics committee. Informed written consent was waived by the local ethics committee (Ethics Committee-Freiburg University Medical Center) due to the purely retrospective analysis. We hereby confirm that all methods were performed in accordance with the relevant guidelines and regulations.All data processing was implemented within our in-house post-processing platform NORA (www.nora-imaging.org; last accessed on 2 January 2022). Preprocessing of diffusion-weighted images included a denoising step [20] followed by correction of the Gibbs-ringing artifacts [21] and final upsampling to isotropic resolution [8].Microstructural diffusion metrics based on a three-compartment diffusion model were estimated using a Bayesian approach [8] to determine the intra-axonal volume fraction (V-intra), the extra-axonal intracellular volume fraction (V-extra), and the extracellular (CSF/free water = V-CSF) volume fraction (Figure 1). T1w imaging datasets were automatically segmented into white matter, grey matter, and cerebrospinal fluid (CSF) using SPM12 (Wellcome Centre for Human Neuroimaging, London, UK). Perilesional T2w white matter hyperintensities were manually segmented on 3D reformatted FLAIR images with co-registration with T1 post-Gd datasets to avoid accidental segmentation of contrast-enhancing tumor components (Figure 2). SPM12 segmentation maps of white matter with the exclusion of T2w-hyperintense areas were defined as whole-brain normal appearing white matter (NAWM). Diffusion metrics were normalized to the NAWM in each case, accounting for age-related white matter changes. To account for potential steroid-related effects, analyses were complemented by excluding patients with corticosteroids, since doses and temporal relation to time interval before imaging were not standardized.Histological analysis of available biopsy tissue followed standardized protocols of the local institute of neuropathology. Sample acquisition followed established diagnostic procedures for fixation in 4% paraformaldehyde, paraffin embedding, and staining. Sections (4 μm thick) were processed and stained with hematoxylin and eosin (H&E). All available biopsy material containing adequate fractions of both tumor tissue and adjacent CNS parenchyma were considered for analysis (available in 6 GBM cases and 6 metastasis cases). Vasogenic edema was assessed on H&E sections by analysis of accentuated tissue disaggregation and increase in translucent areas [22]. Exemplary images were obtained using an Olympus BX40 microscope (Olympus, Shinjuku, Tokyo, Japan) and a Leica DFC450 camera (Leica, Wetzlar, Germany). The samples analyzed in this study had been obtained from surgeries that aimed for radical removal of contrast-enhancing tumor portions, so perilesional T2 changes were not targeted for resection. Therefore, usable histopathologic tissue samples could not be obtained for all tumors studied.Normal distribution assessed by the Shapiro–Wilk test. The Mann–Whitney-U test was employed to compare age and perilesional T2 volumes between GBM and metastasis groups. A One-way ANCOVA, controlling for lesion volume and Bonferroni correction for multiple comparisons was conducted between perilesional T2 areas comparing GBM and metastasis groups. The Pearson’s correlation coefficient was used to relate V-intra, V-extra, and V-CSF to T2 volumes. We plotted the receiver operating characteristic (ROC) curves of GBM and metastasis perilesional and V-extra, V-intra, and V-CSF. Values with an α-level of 0.05 were considered statistically significant. All statistical analyses were performed using R statistics V. 4.0 (R Core Team 2020, Bell Laboratories, Holmdel, NJ, USA; https://www.R-project.org; last accessed 18 February 2022). Boxplots were calculated using CRAN.R-packages (https://CRAN.R-project.org/package=ggplot2 last accessed 18 February 2022, https://CRAN.R-project.org/package=ggstatsplot last accessed 18 February 2022). The Test ROC module was used to calculate ROC curves, which is built on the cutpointR module version 1.1.1 (https://CRAN.R-project.org/package=cutpointr last accessed 18 February 2022).Out of 52 patients presenting with contrast-enhancing intracranial mass lesions, 17 patients (8 female; mean age: 63.5; SD 11.8, range 47.1–85.6 years) with histologically verified brain metastases and 19 patients with GBM (IDH wildtype, 9 female; mean age: 66.4; SD 14.1, range 41.8–88.0 years) underwent presurgical MRI. The following characteristics led to exclusion from the analyses: previous brain surgery (n = 7) and/or radiotherapy (n = 3), IDHmut high-grade gliomas (n = 2; excluded according to the current classification WHO 2021), limited image quality which prevented image postprocessing (n = 4). Corticosteroids had been administered in 7 patients in the GBM group and 6 patients with metastases. Due to the retrospective evaluation, the exact temporal relation of steroid administration to the time of imaging could not be determined. Primary tumors in patients with brain metastases were lung cancer (n = 10), melanoma (n = 3), urothelial carcinoma (n = 1), colorectal carcinoma (n = 1), esophageal carcinoma (n = 1), ovarian cancer (n = 1). The groups did not differ concerning age (p = 0.44) or total volume of tumor-related T2 signaling changes (p = 0.93).The DMI parameters of the T2 hyperintense area were normalized by the values of whole-brain NAWM and compared between GBM and metastasis groups (exemplary in Figure 3), controlling for perilesional T2 hyperintense area volume. Within the perilesional T2 hyperintense area a significant increase in V-CSF in metastases compared to GBM, (F (1,1) = 17.11, p < 0.001), a decrease in V-extra (F (1,1) = 5.60, p = 0.024) and V-intra (F (1,1) = 8.71, p = 0.006) was found. Group-related metrics and ranges are presented in Table 1. The distribution of the individual values is shown in Figure 4. A further analysis excluding patients with corticosteroid treatment led to similar results with an increase in V-CSF in metastases compared to GBM, (F (1,1) = 8.48, p = 0.009) and a decrease in V-extra (F (1,1) = 6.91, p = 0.016) and V-intra (F (1,1) = 9.68, p = 0.005).Results of the Pearson’s correlation indicated a positive association between perilesional T2 volume and V-CSF in metastases (r = 0.51, p < 0.036), without reaching significance in GBM (r = 0.33, p = 0.170). Furthermore, in both groups we found a negative correlation between patient age and perilesional V-CSF (metastases: r = −0.62; p = 0.008; GBM: r = −0.67; p = 0.002).Building on the systematic differences regarding free water content of perilesional T2 hyperintensities between metastases and GBM groups, we conducted a ROC analysis. A model equally weighted for sensitivity and specificity supported the affiliation to the GBM cohort for normalized V-CSF (sensitivity, 63.2%; specificity, 100.0%; PPV, 100.0%; NPV, 70.83%; AUC 0.824) when applying an estimated cutpoint of 3.509. V-intra appeared more sensitive but less specific (sensitivity, 78.95%; specificity, 76.47%; PPV, 78.95%; NPV, 76.47%; AUC 0.752), applying an optimal cutpoint of 0.224. The specificity of V-extra was high but sensitivity low (sensitivity, 42.11%; specificity, 100.0%; PPV, 100.0%; NPV, 60.71%; AUC 0.663) when applying a threshold of 0.989. ROC curves are presented in Figure 5.Histopathological assessment of H&E-stained biopsy material revealed signs of vasogenic edematization, including accentuated tissue loosening and increase in translucent areas in the perilesional parenchyma in all cases with metastases. By contrast, in GBM cases, signs of vasogenic edematization, if present, were unincisive or restricted to small areas of the assessed parenchyma. Exemplary images are shown in Figure 6 (detail histology) and Figure 7 (overview of assessed tumors).Noninvasive differentiation between GBM and brain metastases by conventional MRI remains difficult, as both entities show similar MRI phenotypes. Moreover, in GBM, stratification of non-Gd-enhancing tumor areas is of great clinical importance, as recurrences often occur in these parts of the brain. Assuming that perilesional T2 signal elevations in GBM and metastases differ histopathologically, we used a novel diffusion microstructure imaging approach and examined the obtained image parameters with respect to their diagnostic value.Within perilesional T2 hyperintense areas of metastases, we measured a significant increase in the fraction of free water (V-CSF) compared to the GBM group. This effect persisted even after exclusion of patients pretreated with corticosteroids. Correspondingly, the volume fractions of the intra- and extra-axonal compartments (V-intra and V-extra) were significantly lower in the metastases group compared to the GBM group. Furthermore, a positive correlation between perilesional T2 volume and V-CSF in metastases was observed, which was not the case in GBM. For both brain metastases and GBM groups, a negative correlation between perilesional T2 volume and patient age was found. Concerning the presence of GBM, ROC analysis found high predictive values of both V-CSF (AUC 0.824) and V-intra (0.752) in the perilesional T2 area. The histopathological assessment revealed signs of vasogenic edematization in the perilesional parenchyma of all examined metastases cases, whereas this finding was unincisive or restricted to small areas in the GBM cases.The finding of increased V-CSF in brain metastases fit well to the histopathologically based hypothesis that peritumoral edema around brain metastases is predominantly vasogenic [23], whereas GBM contains additional tumor infiltration [13,24]. While conventional DWI with ADC provides an estimate of the diffusion magnitude, diffusion tensor imaging (DTI) with its FA and MD parameters describes both the magnitude and directionality of diffusion [25]. The advanced diffusion techniques used so far, such as DTI, mostly focused on the FA parameter, which captures the degree of directionality of brain tracts and also correlates with cell density in gliomas. One study showed higher FA levels in the contrast-enhancing tumor fraction in glioblastomas than in brain metastases [26]. Furthermore, in a study that applied five different diffusion techniques to high-grade gliomas and metastases, increased FA values were found in the contrast-attenuated tumor portions in high-grade gliomas compared to metastases [16]. However, none of the applied diffusion techniques succeeded in measuring significant differences in perifocal T2 changes. This could be explained by the fact that the edematous changes would affect the measure of anisotropy and thus the differentiation based on FA and MD is compromised. The 3-compartment model of DMI could in theory allow quantification of distant cell infiltration, but this does not currently seem possible to us for several reasons: on the one hand, reliable differentiation of tumoral and non-tumoral glial cells is not possible, on the other hand, edematization (which also occurs in GBM) is accompanied by a redistribution of cellular compartments and probably also volume increase, which would have an additional impact on the distribution of cellular and noncellular compartments. Therefore, T2 volume was included in the analysis in our study, but it remains unclear to what extent volume effects in GBM influence the distribution of physiological and tumoral cellular structures or the indirect calculation of the parameters V-intra and V-extra. For this, further studies, ideally with quantitative immunohistopathological methods, would be useful.DMI metrics were more widely distributed in the GBM group compared to the metastasis group. This could be explained by infiltrative GBM growth in non-contrast-enhancing areas and possibly higher intertumoral heterogeneity of GBM with varying extent of the vasogenic edema component, whereas in metastases the predominantly vasogenic edema could be more homogeneous at the cellular level.The DMI technique has several potential advantages over NODDI. Violations of the NODDI constraints may induce confounding changes in actually independent parameters, which may lead to less specific results. In contrast, the applied Bayesian approach leads to much softer constraints and better interpretability in pathologically altered tissue. The main contribution to the dMRI signal at the cellular level is defined by microstructural indices such as the diffusion coefficients of the different cell compartments and their volume fractions. However, due to the coarse millimeter resolution of dMRI, much of the microscopic information is usually obscured by mesoscopic changes. The DMI technique disaggregates the microscopic cell properties from the effects of mesoscopic structures. The significantly faster examination time compared to NODDI allows the practical application even in clinical routine MRI examinations.The negative correlation between patient age and T2 volume is difficult to interpret. The extent of neovascularization and vascular endothelial growth factor (VEGF) expression has been shown to influence the development of peritumoral edema in gliomas, meningiomas, and brain metastases [27,28]. Thus, increased VEGF expression is described to be associated with greater peritumoral edema [29], and increased VEGF expression was found more frequently in elderly GBM patients [30]. However, another study in patients with primary GBM did not find an association between age and edema size [31]. Additionally, a selection bias of patient age with the time of examination (which is indirectly related to the expression of the edema) cannot be excluded with absolute certainty. Thus, from our point of view, this finding remains not completely explicable.This study has several limitations: First, the retrospective, monocentric design, the single 3T MRI and the in-house postprocessing pipeline lead to a relatively small number of patients included in this study. Nevertheless, the platform is freely available and further application, also in multicenter cohorts, is possible. Concerning the ROI-based measurements, mean values of both groups were compared and it remains an open question to what extent locoregional heterogeneity exists within the perilesional area. Moreover, diffusion metrics were normalized to whole-brain NAWM. On the one hand, tumor extension in GBM, even beyond the T2 hyperintense parenchyma has been reported in the past [32], on the other hand, we had to account for locoregional variability, which is relevant if normalization ROIs are placed for example within the corticospinal tract. In addition, some of the patients received corticosteroids before the MRI examination. Corticosteroids have been part of the standard treatment of peritumoral vasogenic edema in brain tumors for decades, as they reduce the permeability of tumor capillaries. A reduction in volume of both CE tumor and perilesional T2 changes have been reported for both GBM and metastasis cases [33]. However, we are not aware of exact prospective studies on the effect on quantitative imaging parameters of brain edema.Approximately one-third of cases in both cohorts received corticosteroids and the observed effects did not differ after exclusion of these subgroups of patients. Whether the potentially different effects of corticosteroids on tumor tissue in the respective groups reflects on their properties in MRI, should ideally be investigated in prospective trials with larger patient groups and standardized corticosteroid doses and timing before imaging as the subgroups in our trial were too small to answer this question.To our knowledge, our study is the first to use a novel diffusion microstructure imaging (DMI) sequence and was able to demonstrate significant differences in diffusion metrics between perifocal T2 areas in brain metastases and GBM. The methodology is promising in terms of microstructural characterization of brain tumors and has high diagnostic potential to pre-surgically differentiate GBM from metastases.Diffusion microstructure imaging extends the capabilities of current advanced diffusion imaging techniques by allowing characterization of perilesional white matter T2 changes based on a 3-compartment model. In contrast to GBM, perilesional T2 changes surrounding metastases showed a significant increase in the free water fraction, which was also qualitatively reproduced in histopathological samples. This underscores the concept of a predominant vasogenic-related peritumoral edema around brain metastases, in contrast to perilesional edema and additional infiltrating tumor component in GBM. Thus, diffusion microstructure imaging may not only play an important role in distinguishing GBM from brain metastases but also to further characterize the microstructure of distant tumor components.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051155/s1, Table S1: MRI sequence parameters (3-Tesla MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany).Conceptualization, U.W. and T.D.; methodology, M.R., E.K. and T.D.; software, E.K. and M.R.; validation, U.W. and T.D.; formal analysis, U.W. and T.D.; investigation, U.W. and T.D.; resources, H.U.; data curation, U.W. and T.D.; writing—original draft preparation, U.W.; writing—review and editing, U.W., M.D., D.E., T.D., J.A.H., P.C.R., A.R. and O.S.; visualization, T.D., M.D. and D.E.; supervision, T.D.; project administration, H.U.; funding acquisition, H.U. All authors have read and agreed to the published version of the manuscript. Martin Diebold receives ad-personam funding from the Swiss National Science Foundation and the Bangerter-Rhyner-Foundation. Theo Demerath receives funding from the Research Commission, Faculty of Medicine, University of Freiburg.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board (EK-Freiburg 400/20, approval on 13 August 2020).Patient consent was waived by the local ethics committee in view due to the purely retrospective nature of the study.The anonymized data presented in this study are available on reasonable request from the corresponding author.We acknowledge support by the Open Access Publication Fund of the University of Freiburg. DE was supported by Berta-Ottenstein-Program for Advanced Clinician Scientists, Faculty of Medicine, University of Freiburg. AR was supported by Berta-Ottenstein-Program for Clinician Scientists, Faculty of Medicine, University of Freiburg. The authors would like to thank Konrad Whittaker for proofreading the manuscript. The authors would like to thank Hansjörg Mast for his technical support during the MRI measurements.The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. Peter Reinacher has received research support from Else Kröner-Fresenius Foundation, Fraunhofer Foundation (ATTRACT), German Ministry for Economic Affairs and Energy, and Medical Faculty of the University of Freiburg. He has received personal honoraria for lectures or advice from Boston Scientific, Brainlab, Inomed, and Fraunhofer Foundation and is a consultant to Boston Scientific, Brainlab, and Inomed.MR-physical concept of diffusion microstructure imaging. White-matter electron microscopic (60 nm ultrathin sections) (A) and schematic representation (B). Briefly summarized, the microstructure model can be traced back to a 3-compartment model in which D and v describe the diffusivities and the volume fractions of the corresponding compartments, where the subscript i refers to the intra-axonal compartment, e to the extra-axonal compartment, and f to the free water compartment. Visualization of the extra-axonal and free water compartments are anatomically compromised by preparation artifacts in electron microscopic sections (A). With permission by [19].Axial FLAIR (a) and T1 MPRAGE post-Gd (b) images in a patient with a right occipital brain metastasis with corresponding perilesional FLAIR-hyperintensity (yellow) and contrast-enhancing tumor (blue) segmentations.Axial FLAIR (a,c) and parametric V-CSF-maps (b,d) in a left centroparietal GBM (a,b) and left frontal metastasis (c,d). Of note, the metastasis shows a relative increase in perilesional V-CSF compared with GBM (d,b).Normalized perilesional (T2) diffusion microstructure imaging (DMI) metrics in patients with GBM (n = 19) and metastases (n = 17), normalized to whole-brain normal-appearing white matter (NAWM) values. Compared to GBM, metastases show a significant shift towards increased interstitial free water ((c); V-CSF) and decreased V-intra and V-extra (a,b). * p ≤ 0.05; ** p ≤ 0.01; *** p ≤ 0.001.ROC curves of 19 patients with GBM and 17 with cerebral metastases showing a high predictive value of both perilesional T2 V-CSF (AUC 0.824) and V-intra (AUC 0.752) regarding the presence of a GBM.Example of histological sections in H&E staining in patients with IDHwt GBM (left) and metastasis of ovarian cancer (right). Images in the top row illustrate the immediate neighboring tumor (T) and adjacent parenchymal (P) areas in both samples. Microstructural loosening of the tissue in the surrounding white matter (P) appears prominently in the metastasis case, but is almost absent in glioblastoma, which is particularly visible in detailed images at higher magnification (400×, bottom row), and was interpreted as a sign of increased vasogenic edematization. Magnification bars indicate 200 µm (top row) and 40 µm (bottom row), respectively.Exemplary histopathological sections in 6 patients with IDHwt GBM (upper row) and 6 patients with metastases (lower row). In each case, tumor area (T) and surrounding/infiltrating area (P) are obliquely juxtaposed. Metastases show an increased loosening of the surrounding immediate peritumoral tissue (see also Figure 6) in all cases by visual comparison with GBM. This image impression refers to the immediate peritumoral region. In some cases, the assessability was limited by the retrospective evaluation of surgery-related fragmented tissue. Magnification bars indicate 200 µm.Patient characteristics and ROI (perilesional T2 hyperintense area)-derived diffusion metrics.Values are given in mean and SD or median and interquartile ranges (IQR).Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01156.txt

@@ -0,0 +1,2 @@
+Proton pump inhibitors are frequently used in cancer patients to alleviate some symptoms, epigastric pain or heartburn. However, acid suppression decreases the absorption of some oral-targeted anticancer treatments (tyrosine kinase inhibitors, CDK4/6 inhibitors) and induces changes in the gut microbiome. Recent data are showing that these interactions have important clinical impacts and medical oncologists and patients must be aware of these possible interactions.Multikinase inhibitors (MKIs), and particularly tyrosine kinase inhibitors (TKIs) and immune checkpoint inhibitors (CPIs), are currently some of the major breakthroughs in cancer treatment. Proton pump inhibitors (PPIs) revolutionised the treatment of acid-related diseases, but are frequently overused for epigastric pain or heartburn. However, long-term acid suppression from using PPIs may lead to safety concerns, and could have a greater impact in cancer patients undergoing therapy, like bone fractures, renal toxicities, enteric infections, and micronutrient deficiencies (iron and magnesium). Moreover, acid suppression may also affect the pharmacokinetics of drugs (at least during acid suppression) and decrease the absorption of many molecularly-targeted anticancer therapies, which are mostly weak bases with pH-dependent absorption. This type of drug-drug interaction may have detrimental effects on efficacy, with major clinical impacts described for some orally administrated targeted therapies (erlotinib, gefitinib, pazopanib, palbociclib), and conflicting results with many others, including capecitabine. Furthermore, the long-term use of PPIs results in severe alterations to the gut microbiome and recent retrospective analyses have shown that the benefit of using CPIs was suppressed in patients treated with PPIs. These very expensive drugs are of great importance because of their efficacy. As the use of PPIs is not essential, we must apply the precautionary principle. All these data should encourage medical oncologists to refrain from prescribing PPIs, explaining to patients the risks of interaction in order to prevent inappropriate prescription by another physician.Proton pump inhibitors (PPIs) are one of the most frequently prescribed drugs in the world, and are ranked in the top 10 of US national health-related drug expenditures [1]. These highly efficient drugs in “acid related diseases” are widely available, including “over-the-counter” and at low cost, and are frequently prescribed inappropriately outside of their proven indications (gastric and duodenal ulcer, reflux oesophagitis, prevention of gastrointestinal bleeding when combined with non-steroidal anti-inflammatory drugs, Zollinger–Ellison syndrome) and in long-term use. This overuse is estimated between 40% to 80% in different countries [2,3]. Fortunately, they are very well tolerated, but the initial phase of omeprazole development was stopped when it was shown that carcinoids (ECLome) developed in the oxyntic mucosa in rodents [4]. Nevertheless, in the last decade, growing concerns have emerged regarding their safety, with a large number of studies reporting long-term toxicity, including cancer (of gastric, pancreatic, liver and biliary tract location) [5]. Cancer patients are fragile and many receive long-term PPIs. In a prospective study in four French Comprehensive Cancer Centres, we show that more than a quarter of cancer patients used PPIs, mostly on a daily basis and in the long term [6].Certain side effects of long-term PPI use may be of greater impact in cancer patients than in the general population. On the other hand, long-term suppression of gastric acidity can decrease the absorption, and thus the efficacy, of certain major oral anticancer drugs, as well as changing the composition of the gut microbiome, which also has an impact on the response to immunotherapy [7]. This means that a symptomatic treatment that is not mandatory but is easily removable, might not only produce side effects, but also worsens patients’ prognosis [8,9]. The use of PPIs in cancer patients is thus a real issue [10]. In this review, we aim to update these potential interactions between long-term use of PPIs and cancer patients and their treatment, as well as to propose some possible solutions for cancer patients suffering from heartburn. This question is of particular importance as the use of PPI therapy peaks in older people and cancer predominantly affects the elderly. The biological rationale is based on vitamin B12 deficiency, interaction with certain brain enzymes, and enhanced brain beta-amyloid levels (decreased degradation by lysosomes) [1,11]. A large German prospective cohort study, using observational data, followed more than 73,000 participants over the age of 75 years and free of dementia at baseline. Patients regularly using PPIs (n = 2950) had a significant risk of incident dementia compared with those not using PPIs (HR = 1.44; 95%CI: 1.36–1.52) [12]. Four retrospective and prospective cohorts, however, did not confirm this association [5], which is considered weak [1] when using the Hill criteria (association or causation) [13].It is now widely accepted that PPI use is a risk factor for the development of osteoporosis and osteoporotic fractures [14,15,16]. This can be due to malabsorption of calcium, secondary hyperparathyroidism, and vitamin B12 deficiency. This side effect can be of major importance in the cancer patient population which has accelerated bone loss because of their cancer management [17]. In a population-based cohort, PPI use was associated with a 20–50% higher risk of incident chronic kidney disease, as well as of acute kidney injury [18]. Recently, it has been shown that in patients with chronic kidney disease, chronic use of PPIs accelerates progression of the kidney disease and increases mortality [19]. Another recent retrospective observational study confirmed that in adults with chronic kidney disease, the use of PPIs was associated with an increased risk of hospitalization and mortality [20]. Taking into account the nephrotoxicity (acute but also chronic) of many anticancer drugs, associating them with PPIs should certainly be avoided.The use of PPIs reduces gastric acidity, leading to changes in the gut microbiome, in the same way that antibiotics do. It also decreases colonization resistance to esophageal candidosis and enteric infections including Clostridium difficile, Campylobacter and Salmonella [7]. On a population level, the effect of PPIs on the gut microbiome is more prominent than the effects of antibiotics [21]. These PPI-induced changes in the microbiome may have a clinical impact, particularly in terms of the development of Clostridium difficile infections in the general population. The use of PPIs is also associated with an increased risk of community-acquired pneumonia [22]. In the frail population of cancer patients, long-term PPI prescription may lead to a high risk of enteric infections.Gastrointestinal acidity is important for the absorption of minerals (iron, calcium, magnesium) and vitamin B12. Patients with gastrinoma needing long-term use of high doses of PPIs are a natural model for studying their long-term effects in humans [23]. In this population, long-term use of PPIs was not associated with a decrease in total body stores or iron deficiency [24]. However, in a randomised controlled study in patients with hereditary haemochromatosis, long-term administration of PPIs significantly reduced the volume of blood needed to be removed annually to maintain serum ferritin at 50 µg/L, and 7 days of PPIs significantly decreased absorption of non-haem iron from meat [25,26]. Nevertheless, the development of iron deficit anaemia in patients on long-term PPIs seems infrequent and it is always necessary to exclude other causes. Anaemia in cancer patients often has multiple causes; however, avoiding unnecessary PPIs could be a good policy. Hypomagnesaemia (decreased absorption and increased renal leaks) due to PPIs has been well documented and many dramatic cases have been reported [27]. In 2011, the US FDA released a warning about low serum magnesium levels associated with long-term PPI use. A cross-sectional study in hospitalised patients in Buenos Aires demonstrated that 36% of patients with chronic PPI use had hypomagnesaemia on admission [28]. Association with other drugs used in oncology, and sometimes themselves, the cause of severe hypomagnesaemia, such as cisplatin and EGF receptor antagonists (monoclonal antibodies and tyrosine kinase inhibitors), requires regular follow-up of magnesaemia.The concomitant use of oral antineoplastic agents in patients who are long-term PPI users is a real concern because of the consequences of severe chronic acid suppression, as well as the modifications to the intestinal microbiome. Many papers have addressed the question of the effects of acid suppressive compounds (PPIs and H2 antagonists) on the bioavailability of oral anticancer agents. As TKIs are weakly basic, when the gastric pH is elevated (through the use of PPIs or H2 antagonists) the solubility and bioavailability of these drugs may decrease significantly [29,30]. This decreased bioavailability can sometimes be significant and associated with decreased efficacy. One review reported a major decrease in the oral absorption of crizotinib, dasatinib, erlotinib, gefitinib, lapatinib and pazopanib, and recommended avoiding concomitant use of PPIs or H2 antagonists [31]. A recent systematic review and meta-analysis of the use of gastric-acid suppressants and oral anticancer treatments supports the evidence of a possible negative impact of such combinations on survival outcomes [32].In parallel, there is increasing evidence suggesting that the gut microbiome can modulate the host’s antitumor response and the response to immune checkpoint inhibitors. It has been shown that antibiotics can inhibit the clinical benefits of immune checkpoint inhibitors by modifying the composition of the gut microbiome [33]. PPIs decrease bacterial richness and induce changes in the gut microbiome; these effects are more prominent than the effects of antibiotics [21].TKIs are currently a major weapon in the anticancer arsenal. Oral administration, which is convenient for both patients and physicians, and major efficacy in many forms of cancer, explain why these new drugs are currently one of the major options in the fight against cancer. Most medical oncologists are aware of drug–drug interactions with PPIs, (Table 1) but PPIs are frequently prescribed by the primary care physician, and can even be purchased over the counter, resulting in “unknown” drug–drug interactions that can lead to a decrease in efficacy [34,35].Gefitinib and erlotinib, both selective TKIs targeting the epithelial growth factor receptor, showed reduced absorption in cases of concomitant use with PPIs, [36,37] translating into a significant decrease in efficacy (overall survival and progression-free survival) in retrospective analyses [38,39]. In a large retrospective study of the concomitant use of TKIs and PPIs, nearly 1 in 4 older adults with cancer who received TKIs also received PPIs concomitantly, and this was associated with an increased risk of death—an increase of 21% in lung cancer patients receiving erlotinib and not associated with discontinued use of TKIs [40]. In this study, no impact was observed in the case of co-prescription of PPIs with sunitinib or imatinib, confirming previous results [41]. However, in a real world study, results on the use of PPIs and the impact on first-line sunitinib treatment outcomes are conflicting [41,42]. No impact on serum concentration with PPI use was demonstrated with osimertinib [43].In a retrospective analysis of two prospective trials of pazopanib (one single-arm phase 2, EORTC 62043, and one placebo-controlled phase 3 study, EORTC 62072) in soft-tissue sarcoma patients, of the 333 patients receiving pazopanib, 59 received concomitant PPIs or antiH2; progression-free survival and overall survival were shorter in pazopanib patients receiving gastric antisecretory drugs (respectively 2.8 vs. 4.6 months and 8.0 vs. 12.6 months); these effects were not observed in the placebo group of patients [44]. Clinical pharmacology studies consider that exposure to lenvatinib, vandetanib, cabozantinib, alectinib, and regorafenib is not significantly modified by PPIs [35,45].In hepatocellular carcinoma patients, studies have produced contradictory results; a nationwide cohort study from Taiwan compared patients who took TKIs (sorafenib, regorafenib, lenvatinib and cabozantinib) and were PPI users (n = 2196) with those who were not PPI users (n = 8013). The one-year cumulative incidence of overall mortality was significantly higher in the PPI users (71.3% vs. 61.8%; p < 0.001) and this was confirmed in multiparametric analysis [48]. Similar results were found in a single-centre experience in the UK [49]. However, in secondary analysis of a phase 3 study comparing sorafenib with sunitinib, of the 542 patients receiving sorafenib, 122 were also treated with PPIs at baseline. On univariate and adjusted analyses, no significant association between PPI use and either OS or PFS was identified [50].No known interaction was demonstrated between mTOR inhibitors, PARP inhibitors [51], and PPIs; data regarding BRAF/MEK inhibitors and larotrectinib were scarce but seemed negative [35].The solubility of palbociclib was reduced at pH above 4 and coadministration with PPIs decreased both AUC and Cmax [52]. In metastatic breast cancer patients treated with palbociclib, the concomitant use of PPIs may have a detrimental effect on progression-free survival [53]. On the contrary, gastric pH did not influence the pharmacokinetics of ribociclib.No pharmacokinetic interaction between PPIs and oestrogen receptor inhibitors has been described, but enzalutamide, an androgen receptor inhibitor, can decrease the PPIs’ plasma levels [54]. Recent works on preclinical models, confirmed in retrospective analyses, suggest that patients who received antibiotics around the time of the initiation of immune checkpoint inhibitors (ICI) experienced reduced clinical benefits [33,55]. However, in humans, the effects of PPIs are more prominent than the effects of antibiotics on the gut microbiome [21]. Numerous studies have thus addressed the problem of ICI efficacy in PPI users.In a cohort of 112 melanoma patients treated with anti PD-1, significant differences were observed in the microbiomes of responders versus non-responders [56]. In a retrospective analysis from CheckMate 069, the objective response rate (and PFS) after immunotherapy (ipilimumab alone or combined with nivolumab) in patients receiving PPIs was half that of those not on PPIs [57]. In 2020, retrospective analysis using pooled data from one phase 2 and one phase 3 trial comparing atezolizumab (n = 757) with docetaxel (n = 755) in previously-treated non-small-cell- lung cancer (the POPLAR and OAK trials) showed that PPI use was associated with shorter OS and PFS in the atezolizumab population and not in the docetaxel population [58]. Individual participant data from two urothelial cancer trials (IMvigor210 and 211) testing the efficacy of atezolizumab were analysed retrospectively with regard to the concomitant use of PPIs (approximately 30% of patients). In the pooled group of patients receiving atezolizumab (n = 847), PPI use was a negative prognostic marker (for overall survival, progression-free survival and response rate); in the randomised trial, atezolizumab showed significant efficacy on OS versus chemotherapy (HR: 0.69; 95% CI: 0.56–0.84) for PPI non-users and no OS benefit (HR: 1.04; 95% CI: 0.81–1.34) for PPI users; the same results were observed for PFS and ORR [59]. The phase 3 trial, IMpower 150, compared in non-small cell lung cancers, three chemotherapy regimens, two composed of atezolizumab. In post hoc analysis (1202 participants, 441 receiving PPIs), PPIs use was independently associated with worse overall survival in the pooled atezolizumab arms (n = 748), but not in the third arm without ICI [60]. The OS effect of atezolizumab was negative for PPIs users (HR: 1.03; 95% CI: 0.77–1.36), while it was clearly positive for non-users (HR: 0.68; 95% CI: 0.54–0.86). The concomitant use of PPIs thus transforms a major breakthrough drug into a treatment that is inefficient. (Table 2).In a recent Korean cohort study of 2963 patients treated with ICIs as the second line, for non-small cell lung cancer, 936 were concomitant PPIs users. After propensity score matching (1:1 ratio), 1646 were analysed. The use of PPIs was associated with a higher risk of mortality compared to non-use (HR: 1.28; 95% CI: 1.13–1.46) [61].An Italian series evaluated the prognostic impact of concomitant treatments (antibiotics, PPIs, or corticosteroids), quantified by a drug score, in a large series of patients receiving pembrolizumab or chemotherapy for non-small cell lung cancer. This drug score had a predictive value for response rate, OS and PFS, essentially in the pembrolizumab cohort [62].Recently, a meta-analysis of seven studies (3647 cancer patients) was reported. The authors concluded that PPIs’ use had a detrimental effect on the efficacy of ICI: PPIs’ use increased the risk of death by 39% and the risk of progression by 28% [63].In Bordeaux University Hospital, between May 2015 and September 2017, 635 patients received CPI for cancer. The authors analysed the influence of comedications (including PPIs) on the anti-tumour effect and safety of these CPI. PPIs were prescribed in 38% of these patients; the median OS of patients receiving PPIs was 9 months versus 26.5 months in those not receiving PPIs (HR: 1.70, 95%CI: 1.40–2.08) [64].High doses of parenteral methotrexate are used in some forms of cancer and require strict drug monitoring. In a series of 74 patients receiving high dose methotrexate, it was demonstrated that co-administration of PPIs was associated with delayed elimination of methotrexate, as well as renal and liver dysfunction [65]. The mechanism is uncertain, but PPIs should be used cautiously with a high dose of methotrexate. In 2017, secondary unplanned analysis of the TRIO-013 trial comparing capecitabine and oxaliplatin (CapOx) with or without lapatinib in ERB2-positive metastatic gastroesophageal cancer aimed to determine if orally administered capecitabine or lapatinib were hampered by concomitant prescription of PPIs [66]. Of the 545 randomised patients, 229 (42%) evenly distributed patients received PPIs. In the placebo arm (receiving CapOx only), patients treated with PPIs had worse efficacy results (PFS, disease control rate, and OS) than those not receiving PPIs. The same authors conducted retrospective analysis of patients with stage II–III colorectal cancer who received adjuvant CapOx or FOLFOX in Edmonton, Alberta. Between 2004 and 2013, 389 patients, 214 receiving CapeOx and 175 receiving FOLFOX, met their inclusion criteria; respectively, 50 (23.4%) and 49 (28%) had concomitant PPIs. Three-year recurrence-free survival was significantly lower in the CapeOx-treated PPIs recipients than the non-PPIs recipients. This was not demonstrated in the FOLFOX-treated PPI recipients, but the differences were minor [67]. More recently, secondary analysis of six randomised controlled trials in patients with advanced colorectal cancer treated with fluoropyrimidines was conducted using individual patient data. Of the 5594 patients included, 902 received PPIs at trial entry. PPIs’ use was significantly associated with worse overall survival (pooled HR, 1.20; 95% CI, 1.03–1.40; p = 0.02) and progression-free survival (overall pooled HR, 1.20; 965% CI: 1.05–1.37; p = 0.009); this was particularly obvious for patients under 5FU and not among those receiving capecitabine; nor was it obvious for patients treated with other gastric antisecretory drugs (such as H2 antagonists). The authors concluded that clinicians should cautiously consider the concomitant use of PPIs in such patients. The mechanistic basis was unclear: impact on several transporters, modifications to intracellular pH, or something else [68]. Future studies are thus warranted as a series are accumulating on such possible interactions [69].To conclude, the effect of PPIs on the efficacy of certain anticancer agents, particularly TKIs and CPIs, is a major issue in daily practice. In this opinion paper, we have put emphasis on articles showing the potential negative impact of such combinations and particularly on unplanned retrospective analysis from prospective studies, because we can expect that no randomized trial can be and will be conducted on this topic; moreover, PPIs are symptomatic treatments that can be replaced without any major risk of interactions. There are articles that did not find clinical interactions, particularly with CPIs [70,71,72], but we think that the precautionary principle must be applied until there is demonstration of the absence of clinical interaction. It is certainly of major importance that patients can be helped to stop taking PPIs after four weeks of treatment, except in cases of severe oesophagitis, previous bleeding, or Barrett’s oesophagus, [73] and ideally that prescriptions of PPIs be avoided for heartburn or epigastralgia. Some tricks, such as drinking acidic beverages (cola) with erlotinib could be proposed, but the best way is certainly to replace these long-lasting drugs with other therapeutic means [74]. If the use of acid-suppressive drugs is necessary, H2 antagonists (ranitidine) can be used and given 2 h after TKIs. Antacids can also be used 2 h before or after the drug [31]. The use of PPIs should be limited to TKIs with no proven interactions between absorption and intragastric pH. In patients treated with CPIs, the interaction is not due to drug absorption but rather to the alteration of the gut microbiome and we can suppose that the negative effect may also be observed after long-term use of H2 antagonists. In such cases, antacids are the best option, although on-demand use of PPIs or H2 antagonists may also be proposed.Conceptualization, J.-L.R. and J.-S.F.; validation, all authors.; writing—original draft preparation, J.-L.R.; writing—review and editing, J.-L.R., J.E., M.G.; visualization, all authors.; supervision, V.S., C.M.-B. All authors have read and agreed to the published version of the manuscript.The APC was funded by Institut de Cancérologie de l’Ouest.The authors declare no conflict of interest. Pharmacokinetic (PK) interactions between H2 antagonists (H2A) or proton pump inhibitors (PPI) and tyrosine kinase inhibitors; recommendations and demonstration of the clinical impact of such interactions.Pharmacokinetic interactions: NA: no data available; 0: definitively no interactions; ±:
+			conflicting results; +: possible interactions; ++: clear interactions; +++: major interactions. Recommendations: NA: no data available, OK: concomitant use possible; no: concomitant use strongly discouraged. Clinical impact of concomitant use: NA: no data available; YES: clinical impact demonstrated in clinical series; Conflicting results: clinical series showing different results.Overall survival results of 2 randomised controlled studies comparing atezolizumab vs. systemic chemotherapy with proton pump inhibitor (PPI) users versus non-users. HR: hazard ratio of overall survival of atezolizumab versus chemotherapy.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01157.txt

@@ -0,0 +1 @@
+The comparative effectiveness and safety of the standard dose and lower doses of pembrolizumab in non-small-cell lung cancer (NSCLC) patients still remains limited. We conducted a retrospective multi-institutional cohort study of patients newly initiating pembrolizumab in Taiwan. We found that the median overall survival (OS) and rate for all classes of immune-related adverse events (irAEs) were similar for both the standard-dose and low-dose pembrolizumab groups. Moreover, we found that patients with a pembrolizumab dose ≥1.8 mg/kg were associated with better OS than those receiving <1.8 mg/kg. Our findings suggested that a pembrolizumab dose ≥1.8 mg/kg may be the clinically minimally efficient dose.Fixed doses at 200 mg of pembrolizumab or 2 mg/kg every 3 weeks are the standard dosages for first- and second-line treatment of non-small-cell lung cancer (NSCLC); however, in clinical practice, patients with NSCLC may receive lower doses of pembrolizumab due to drug product availability or economic factors. To date, the comparative effectiveness and safety of the standard dose and lower doses of pembrolizumab in these patients still remains limited. We conducted a retrospective cohort study by analyzing electronic medical records data from the largest multi-institutional hospital system in Taiwan. Advanced NSCLC patients newly receiving pembrolizumab with or without chemotherapy were included. Patients were classified into: (1) the standard-dose group (≥2 mg/kg), and (2) the low-dose group (<2 mg/kg). We applied inverse probability of treatment weighting (IPTW) to compare the overall survival (OS) and immune-related adverse events (irAEs) between the two treatment groups, and to evaluate the minimum clinically effective dose of pembrolizumab. We included a total of 147 NSCLC patients receiving standard-dose pembrolizumab (mean [range] age: 63.7 [58.0–73.0] years; male: 62.6%; mean [range] body weight: 60.5 [58.0–73.0] kg) and 95 patients receiving low-dose pembrolizumab (mean [range] age: 62.0 [50.0–68.8] years; male: 64.2%; mean [range] body weight: 63.9 [55.0–73.8] kg). After IPTW adjustments, the median OS was similar for both the standard-dose and low-dose pembrolizumab groups (19.3 vs. 14.3 months, log-rank p = 0.15). Also, the rate for all classes of irAEs was similar for both groups. We found that patients with a pembrolizumab dose ≥1.8 mg/kg were associated with better OS than those receiving <1.8 mg/kg. Our findings suggested no significant difference in OS and irAEs between patients receiving pembrolizumab ≥2 mg/kg and <2 mg/kg in clinical practice. A pembrolizumab dose ≥1.8 mg/kg may be the clinically most efficient dose.Lung cancer is the leading cause of cancer-related mortality and causes significant health and economic burden worldwide [1]. Non-small-cell lung cancer (NSCLC) is the most frequent (85–90%) cause of lung tumors [2]. Previously, chemotherapy and tyrosine-kinase inhibitors (TKI), based on the presence of oncogenic driver mutations, were the standard treatments for NSCLC patients [3]. However, several breakthrough novel agents have been developed for advanced NSCLC therapy which have prolonged survival for those patients. For example, antibodies of anti-program-death 1 (PD-1) or anti-program death-ligand 1 (PD-L1) (i.e., nivolumab and pembrolizumab) can disrupt the interaction between PD-1 and PD-L1, an interaction which inactivates T cells and thus permits cancer cells to evade the human immune system [4]. These novel agents have now become the first-line treatment for advanced NSCLC [3,5].The phase 1 trial of KEYNOTE-001 designated the dosage of pembrolizumab at 2 mg/kg every 3 weeks as the standard dose for second-line treatment in NSCLC patients. However, subsequent clinical trials, including KEYNOTE-024 [6,7] and KEYNOTE-042 [8,9], chose a fixed dose of pembrolizumab 200 mg every 3 weeks as first-line treatment for NSCLC patients. Model-based study has shown that saturation in ex vivo target engagement is similar for the 2 mg/kg dose and the 200 mg fixed dose, both at 3-week intervals [10,11]. Moreover, the comparative data of pharmacokinetics, toxicity and clinical responses shows no significant differences between the weight-based dose (2 mg/kg) and the fixed-dose regimen (200 mg) of pembrolizumab [12]. The fixed-dose regimen offers greater convenience in clinical practice, but the weight-based dose regimen may offer a 24–26% cost savings for NSCLC treatment with pembrolizumab [13,14,15].Due to the economic burden of treatment, patients may only be able to afford single 100 mg vials of pembrolizumab, the minimum strength of the drug, to treat NSCLC every 3 weeks. However, a fixed dose of 100 mg of pembrolizumab is only appropriate for patients weighing less than 50 kg, following the weight-based dosing of 2 mg/kg during each treatment course. Recent simulation data have shown that a lower dose of pembrolizumab at 1 mg/kg also provides adequate trough target engagement (with occupancy of 96.8% for patients weighing 70 kg) [10]. Freshwater, T. et al. also reported similar exposure distributions between the standard dose (≥2 mg/kg) and low dose (<2 mg/kg) of pembrolizumab [11]. To date, the evidence of treatment outcomes from different weight-based doses of pembrolizumab in NSCLC patients is still lacking; we have therefore harnessed multi-institutional electronic medical records in Taiwan to compare the effectiveness and safety between standard-dose and low-dose pembrolizumab for NSCLC in real-world clinical practice. Because the ex vivo pharmacokinetics evidence might not totally reflect the real-world pharmacodynamics evidence, our secondary objective was to explore the real-world minimum effective dose of pembrolizumab.The Chang Gung Research Database (CGRD) contains the electronic medical records (EMR) of patients from eight Chang Gung Memorial Hospitals (CGMH) and represents about 14.0% of all cancer patients in Taiwan [16]. The CGRD contains records of individual-level demographics, health conditions, all medications records, laboratory examination results, pathological findings, radiologic results, and medical procedures. The database profiles have been described in previous literature [16]. In this study, we linked CGRD to the Taiwan Cancer Registry and the National Cause of Death Registry to obtain more details on cancer stage, cancer treatments and death outside CGMH. The data quality and validity in this dataset have been documented [17,18,19]. This study has been approved by the Institutional Review Board of the Chang Gung Medical Foundation (202001612B0C601).We identified all patients diagnosed with stage Ⅲ/Ⅳ NSCLC (International Classification of Diseases, Tenth version (ICD-10) disease code C34.0–C34.9) during 2015–2020 from CGRD. Patients diagnosed with other cancers were excluded because of different disease prognostic levels. We included patients newly receiving at least 2 treatment courses of pembrolizumab with or without chemotherapy between 1 January 2016 and 31 December 2019. The first date of pembrolizumab treatment was defined as the index date. The details of study cohort selection are shown in Figure 1.Patients eligible for the study were divided into 2 groups depending on the weight-based pembrolizumab dose they received: (1) the standard-dose group (≥2 mg/kg), and (2) the low-dose group (<2 mg/kg). The patients’ weight and pembrolizumab doses for NSCLC were recorded on the index date. Pursuant to our secondary objective to explore minimum dosing, we redefined the standard group as ≥1.9 mg/kg (dosing rounded down 5%) and ≥1.8 mg/kg (dosing rounded down 10%), following a dose band strategy (10% variance) [20]. We recorded baseline information regarding patients’ characteristics such as age, Eastern Cooperative Oncology Group (ECOG) performance status, stage, histologic features, metastasis status and PD-L1 tumor proportion score on the index date. Patient comorbidities and co-medication such as alcohol behavior, smoking behavior, Charlson comorbidity index (CCI), chronic diseases and biochemical data were collected from one year prior to the index date.We followed up the eligible patients from the index date to patient death, last clinical visit, loss of follow-up or 31 December 2020, whichever came first. The primary effectiveness outcome was overall survival (OS), which was defined as the time from index date to death. The secondary effectiveness outcome was the time to tumor progression (TTP), according to Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1. The safety outcomes included serious adverse events post pembrolizumab treatment, defined as requiring intravenous steroid treatments for more than 3 days, and any grade immune-related adverse events (irAEs) (i.e., skin irAEs, hepatic irAEs and endocrine irAEs) according to the National Cancer Institute Common Toxicity Criteria for Adverse Events (CTCAE) version 4.0. The TTP and safety outcome assessments were judged by the multidisciplinary lung cancer conference and recorded on patients’ EMR. Detailed definitions of irAEs are shown in Supplemental Table S1.We described the patients’ characteristics by mean and interquartile range (IQR) for continuous variables, and absolute and relative frequencies for categorical variables, respectively. We compared the patients’ characteristics between the standard and low-dose pembrolizumab groups using Wilcoxon or Chi-square tests. To make the two treatment groups more homogeneous, we applied a propensity score (PS) using inverse probability of treatment weighting (IPTW) [21]. In the IPTW method, the low-dose pembrolizumab group was weighted by the inverse of the estimated PS and the standard-dose pembrolizumab group was weighted by the inverse of 1 minus the estimated PS. The Kaplan–Meier (KM) method and Cox regression model adjusted by IPTW were performed to compare the median OS and hazard ratio (HR) with 95% confidence intervals (95% CI) between the standard- and low-dose pembrolizumab groups. Moreover, we performed subgroup analyses by clinically important characteristics, including sex, age (<65 years, ≥65 years), smoking, weight (<50 kg, ≥50 kg), PD-L1 score (<50%, ≥50%), brain metastasis, concomitant chemotherapy, line of pembrolizumab therapy and experience of skin irAEs. Additionally, we also performed post hoc subgroup analyses of all included patients by line of pembrolizumab therapy, concomitant chemotherapy, experienced skin irAEs and a fixed dose of pembrolizumab (100 mg vs. 200 mg). Specifically, we reperformed the IPTW methods to generate more homogeneous comparisons between the standard- and low-dose pembrolizumab groups in each subgroup analysis. After the IPTW adjustments, we would further control the imbalanced baseline characteristics to measure the HR by using multi-variable Cox proportional hazard models. We performed all analyses using SAS Enterprise Guide, version 7.1 (SAS Institute, Cary, NC, USA). The details of SAS programs are described in Supplemental Table S2.A total of 242 NSCLC patients newly receiving pembrolizumab were included in our study. The patients’ characteristics are presented in Table 1. Of the patients, 63.2% were male with a mean age of 62.0 years (range: 56.0–72.0) and weight of 61.5 kg (52.8–69.4). More than half of the patients received pembrolizumab as first-line therapy (63.6%) for advanced NSCLC and 54.5% of the patients combined pembrolizumab with chemotherapy. Most patients had ECOG performance status less than 2 (86.8%). Among patients’ tumor histological features, 67.7% were classified as adenocarcinoma. Before pembrolizumab treatment, 114 (47.1%) had a PD-L1 tumor proportion score more than 50%.The standard-dose group (pembrolizumab ≥2 mg/kg) consisted of 147 (60.7%) patients, and the low-dose group (pembrolizumab <2 mg/kg) consisted of the remaining 95 patients. Before the IPTW adjustment, the tumor burden, baseline characteristics and biochemical data were similar between the two groups (Table 1). However, the standard-dose group had a lower body weight (mean: 60.5 vs. 63.9 kg, p = 0.04) and level of alanine aminotransferase (ALT) (28.6 vs. 29.7 U/L, p = 0.01) than the low-dose group. Patients in the standard-dose group had a higher proportion of chronic obstructive pulmonary disease (COPD) (26.5% vs. 13.6%, p = 0.01) and a higher level of white blood cells (WBC) (8.4 vs. 7.6 103/uL, p = 0.02). All baseline characteristics were comparable between the two groups after IPTW (Table 1).Among these pembrolizumab new users, 107 (44.2%) patients died during the median follow-up of 10.1 months (range: 4.4–17.3 months). The median follow-up times for the standard- and low-dose groups were 9.9 and 10.6 months, respectively. After IPTW adjustment, the median OS was longer in the standard-dose group but did not reach statistical significance compared with the low-dose group (19.3 vs. 14.3 months, log-rank p = 0.15) (Figure 2a). The median TTP was similar in the standard- and low-dose groups (3.4 vs. 2.8 months, log-rank p = 0.12) (Figure 2b). For our secondary analyses, we adjusted the 2 mg/kg pembrolizumab dosing by rounding down by 5% and 10%, and we presented the baseline characteristics of redefining the standard- and low-dose groups after IPTW adjustment in Supplemental Tables S3 and S4. Taking pembrolizumab ≥1.8 mg/kg as the new standard-dose group, we found significantly better OS in the standard-dose group than in the low-dose group (HR: 0.73, 95% CI: 0.55–0.97) (Table 2). Rates for all classes of irAEs are summarized in Supplemental Table S5. The irAEs rate was similar in both groups.In the sensitivity analyses, we applied propensity score matching to make the two groups more comparable, and the result was consistent with the IPTW pseudo-population (HR: 0.83, 95% CI: 0.52–1.34). The primary, secondary and sensitivity analyses of OS and TTP are summarized in Table 2, Figure 2 and Supplemental Figure S1. The better OS of the standard-dose group in all subgroup analyses, while not reaching statistical significance, was similar to that of the overall population; however, statistically significant OS improvement was found in several group analyses (Figure 3), including age ≥65 years (HR: 0.63, 95% CI: 0.40–1.00), non-smokers (HR: 0.62, 95% CI: 0.43–0.88), weight ≥50 kg (HR: 0.68, 95% CI: 0.48–0.95), no brain metastasis (HR: 0.72, 95% CI: 0.53–0.98), no combination with chemotherapy (HR: 0.57, 95% CI: 0.38–0.87) and first-line pembrolizumab therapy (HR: 0.64, 95% CI: 0.43–0.95).Post hoc subgroup analyses showed that OS was significantly better in patients receiving pembrolizumab as first-line therapy (median OS: 25.6 vs. 12.0 months; p < 0.01) (Figure 4a), those with skin irAEs (median OS: 25.6 vs. 13.6 months; p < 0.01) (Figure 4b) and those receiving fixed-dose 200 mg pembrolizumab (median OS: non–reach vs. 12.5 months; p < 0.05) (Figure 4d). However, the combination of chemotherapy and pembrolizumab showed similar effectiveness compared with pembrolizumab monotherapy (median OS: 18.0 vs. 17.6 months; p = 0.50) (Figure 4c).Based on Taiwan’s largest multi-institutional EMR database, our retrospective study showed no significant clinical benefits and safety difference between the standard-dose group (≥2 mg/kg) and the low-dose group (<2 mg/kg). Moreover, we found the minimum effective dose of pembrolizumab to be 1.8 mg/kg, meaning that 55.5 kg NSCLC patients could be treated with only one single 100-mg pembrolizumab vial. In our subgroup analysis, the standard dose was associated with better OS in patients aged 65 or above, non-smokers, those receiving pembrolizumab for first-line use in advanced NSCLC, those not combining with chemotherapy and those without brain metastasis.Few real-world studies have compared clinical outcomes between standard-dose and low-dose pembrolizumab for NSCLC patients. A previous single-center study from Low, J.L. et al. in Singapore reported no significant differences in median OS between two groups receiving ≥2 mg/kg and <2 mg/kg (13.5 vs. 14.7 months) [22]. Also, pharmacokinetic studies have reported a 95% trough target engagement with dosing at 0.8 mg/kg every three weeks and saturation of PD-L1 receptors at a dose of ≥1 mg/kg [10,11,22]. Furthermore, the present study showed comparable median OS between the standard-dose and low-dose groups (19.3 vs. 14.3 months). It seems that some of the NSCLC patients who received low-dose pembrolizumab might be deriving clinical benefits at the low dose.Rate of irAEs in our study were slightly higher than those reported in the pivotal phase 3 studies [6,7,23,24] and similar to those reported in the previous real-world study [25]. More frequent contact and intensive education, and highly selected patients in the well-controlled trial setting may contribute this discrepancy. For example, poorer performance status in our and previous real-world study population than previous trials’ population might have a higher incidence of irAEs. Nevertheless, rate of irAEs were not significantly different between the standard-dose and low-dose groups and we suggested that healthcare providers should monitor the occurrence of irAEs regardless of dosage.The Low. J.L. et al. study also showed better median OS but no significant difference between fixed-dose 200 mg and 100 mg (19.8 vs. 14.3 months) [22]. However, our study showed that a fixed-dose 200 mg of pembrolizumab was associated with significantly better OS than fixed-dose 100 mg (non-reach vs. 12.5 months). Previous systematic reviews showed that PD-L1 expression was associated with better OS for NSCLC patients under pembrolizumab treatment [26,27]. Hence, comparing patient data of our study with those of Low, J.L. et al., we found that 68% and 39%, respectively, of the fixed-dose 200 mg patients had PD-L1 expression ≥50%, and 43% and 68%, respectively, of the fixed-dose 100 mg patients had PD-L1 expression ≥50%. Although the percentage of PD-L1 expression ≥50% between fixed-dose 200 mg and 100 mg did not reach the statistical significance (68% vs. 43%, p = 0.06) in our study, it might be an important factor that contributed different median OS. Otherwise, evidence directly comparing pembrolizumab alone versus pembrolizumab combined with chemotherapy for NSCLC patients remains scant. Previous indirect comparison evidence has shown controversial outcomes in OS and improved outcomes in progression-free survival (PFS) [28,29,30]. Again comparing our study with Low, J.L. et al., 77% and 35% of patients, respectively, received fixed-dose 200 mg pembrolizumab alone, while 40% and 74%, respectively, received fixed-dose 100 mg pembrolizumab alone. Details of baseline characteristics of our study and the Low, J.L. et al. study between fixed-dose 200 mg and the 100 mg group are summarized in Supplemental Table S6. Differences in demographics (i.e., PD-L1 expression) and treatment patterns between countries may give rise to the observed differences in OS between the two groups (200 mg and 100 mg fixed-dose groups).To our knowledge, there has been no analysis based on real-world evidence to determine the minimum clinically effective dose of pembrolizumab for NSCLC treatment. A model-based study from the KEYNOTE–001 trial showed saturation of ex vivo target engagement in blood beginning at a dose of ≥1 mg/kg every 3 weeks [10]. Our study tried to determine the lowest cutoff value based on real-world clinical practice, and we found that overall survival differed significantly between the weight-based dose ≥1.8 mg/kg and <1.8 mg/kg. Asian real-world evidence showed that the median pembrolizumab dose was 1.85 mg/kg (range: 1.24–3.2 mg/kg every 3 weeks), with their patients receiving a fixed dose at 100 mg2. Due to the small sample size, they did not analyze clinical outcomes for the lowest-dose cut-off value. Our study provides evidence that the ex vivo pharmacokinetics models may not fully apply to real-world clinical practice. However, due to the limitations of our retrospective study design, pharmacokinetics data (i.e., blood concentration of pembrolizumab) was not available to allow a comparison with previous model-based studies [10,11].In our subgroup analyses, standard-dose groups were associated with significantly better OS in some predefined subcohorts. In previous randomized controlled trials, the clinical benefits of immune checkpoint inhibitors (ICIs) were not significantly better in patients aged 65 or over, and nonsmokers, as compared to placebo or chemotherapy. Recent meta-analyses, pooling all published randomized trials of standard-dose ICIs, showed consistent clinical benefits among elderly and non-elderly patients [31]. However, ICIs might be less effective in nonsmokers at the standard dose, and lower dosing of ICIs might worsen clinical outcomes. Moreover, patients receiving pembrolizumab treatment alone and those receiving it as first-line therapy had better OS at the standard dose. These findings correspond to the pivotal phase 3 study and current treatment guidelines [3,6,7]. Taking together our findings and previous literature, we suggest that patients aged ≥65 and nonsmoking patients should receive the standard dose of pembrolizumab for the treatment of NSCLC.Our study provides real-world evidence from Taiwan on different doses of pembrolizumab therapy for advanced NSCLC. However, several limitations should be noted before interpreting the results. First, due to the inherent nature of retrospective designs, possible impacts from confounding factors must be acknowledged. To mitigate this, we applied IPTW and PS matching methods to make the comparisons more homogeneous. Second, patients may become lost to follow-up in the study hospitals. Therefore, we linked our EMR database to the national death records database to capture any deaths outside the study hospitals, as a result of which we consider our estimates of median OS unbiased [32]. Third, incidence of irAEs might be underestimated due to some grade 1 adverse events only recorded in unstructured medical records. NSCLC patients receiving pembrolizumab with irAEs had better clinical outcomes compared to those without irAEs [25,33]. However, we defined irAEs using drug records and laboratory examinations and found that the incidence of irAEs was similar for the two groups. We therefore consider that incidence of irAEs is unlikely to have been underestimated.This multi-institutional cohort study suggested no significant differences in OS and irAEs between NSCLC patients receiving pembrolizumab ≥2 mg/kg (standard-dose group) and <2 mg/kg (low-dose group) in Taiwan. NSCLC patients with pembrolizumab dose ≥1.8 mg/kg were associated with better OS than those receiving <1.8 mg/kg. Future studies on the effectiveness and safety from the lower dose of pembrolizumab are suggested to replicate our findings.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051157/s1, Figure S1: Kaplan-Meier curve of OS and TTP between standard-dose and low-dose groups before IPTW adjustment; Table S1: Definition of immune-related adverse events, Table S2: SAS programs for the data analyses in this study, Table S3: Baseline characteristics (re-defined the standard group as ≥1.9 mg/kg), Table S4: Baseline characteristics (re-defined the standard group as ≥1.8 mg/kg), Table S5: Summary of immune-related adverse events, Table S6: Baseline characteristics of our study and Low J.L. et al. study between fixed-dose 200 mg and 100 mg group.Conception and design: K.-C.C., S.-C.S., Y.-F.F.; analysis and interpretation of the data: K.-C.C., S.-C.S.; drafting of the article: K.-C.C., S.-C.S.; critical revision of the article for important intellectual content: all authors; final approval of the article: all authors; provision of study materials or patients: H.-Y.C., Y.-Y.C.; statistical expertise: K.-C.C., S.-C.S.; administrative, technical, or logistic support: H.-Y.C., Y.-Y.C., Y.-F.F.; collection and assembly of data: K.-C.C., S.-C.S. All authors have read and agreed to the published version of the manuscript.This study was funded in part by the following: grants from the Ministry of Science and Technology of Taiwan (110–2314–B–182A–138–MY3 to Y.F. Fang) and Chang Gung Memorial Hospital (Grant CMRPG3J1313 to Y.F. Fang).The study was conducted in accordance with the Declaration of Helsinki, and approved by the Institutional Review Board of the Chang Gung Medical Foundation (No.: 202001612B0C601, approval on 20 November 2020).Patient consent was waived due to the retrospective nature of this study.The data presented in this study are available on request from the corresponding author.The authors wish to acknowledge the support of the Maintenance Project of the Center for Big Data Analytics and Statistics (Application number: 00498–2020111787856) at Chang Gung Memorial Hospital and express their gratitude for database maintenance and statistical assistance.The authors declare no conflict of interest.Selection of study cohort from database.Kaplan–Meier curve of OS and TTP between standard-dose and low-dose groups after IPTW adjustment. (a) OS between standard-dose group and lower-dose group; (b) TTP between standard-dose group and lower-dose group cohort; (c) OS between standard-dose group (≥1.9 mg/kg) and lower-dose group; (d) TTP between standard-dose group (≥1.9 mg/kg) and lower-dose group; (e) OS between standard-dose group (≥1.8 mg/kg) and lower-dose group; (f) TTP between standard-dose group (≥1.8 mg/kg) and lower-dose group. Note: * The patient number was adjusted by IPTW methods.Subgroup analysis of overall survival after IPTW adjustment.Post hoc subgroup analyses of weighted Kaplan–Meier survival curve by clinically specific group after IPTW adjustment. (a) Pembrolizumab first line vs. second line and more; (b) Skin irAEs vs. Non–skin irAEs; (c) Pembrolizumab alone vs. combined chemotherapy; (d) Pembrolizumab fixed dose 100 mg vs. 200 mg.Baseline characteristics.Abbreviations—CCI: Charlson comorbidity index, ECOG: Eastern Cooperative Oncology Group, PD-L1: programmed death-ligand 1, COPD: chronic obstructive pulmonary disease, eGFR: estimated glomerular filtration rate, ALT: alanine aminotransferase, AST: aspartate aminotransferase, TSH: thyroid-stimulating hormone, WBC: white blood cell. Note: Continuous variables are expressed as mean (Q1–Q3) and dichotomous variables are expressed as percentage (%). * Since the patients were adjusted by IPTW, the total patient number was not identical to that in the original cohort.Sensitivity analyses of overall survival and time to tumor progression between standard-dose and low-dose groups after IPTW adjustment.Abbreviations—HR: hazard ratio, CI: confidence interval: Note: † adjusted by inverse probability of treatment weighting; * p < 0.05; Propensity score model covariates include: age, sex, Eastern Cooperative Oncology Group, alcohol habit, smoking habit, line of pembrolizumab, concomitant chemotherapy, Charlson comorbidity index, histologic features, metastasis status, programmed death-ligand 1 tumor proportion score and chronic diseases.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01158.txt

@@ -0,0 +1 @@
+Absence of survival benefits when adding hyperthermic intraperitoneal chemotherapy (HIPEC) with oxaliplatin to cytoreductive surgery in peritoneal metastasis from colorectal cancer has recently been shown in the randomized controlled PRODIGE 7 trial. We therefore aimed to investigate the effects of this treatment modality in a preclinical micrometastasis model. Cancer cells were incubated with either patient samples obtained during HIPEC procedures or with defined oxaliplatin-containing solutions prepared according to clinically established HIPEC protocols. Our results demonstrate a limited effectiveness of short-term HIPEC in simulations with oxaliplatin to eliminate micrometastases, although we used platinum-sensitive cell lines for our model. Since these results are in line with findings from current research, our studies might offer further convincing evidence and potential explanations for HIPEC futility observed in clinical application.Cytoreductive surgery combined with hyperthermic intraperitoneal chemotherapy (HIPEC) was considered a promising treatment for patients with peritoneal metastasis from colorectal cancer. However, the recently published randomized controlled PRODIGE 7 trial failed to demonstrate survival benefits through the addition of short-term oxaliplatin-based HIPEC. Constituting a complex multifactorial treatment, we investigated HIPEC in a preclinical model concerning the elimination of minimal tumor residues, thereby aiming to better understand the size of effects and respective clinical trial results. Patient samples of peritoneal perfusates obtained during HIPEC treatments and oxaliplatin-containing solutions at clinically relevant dosages, conforming with established HIPEC protocols, were assessed regarding their ability to eliminate modelled ~100 µm thickness cancer cell layers. Impedance-based real-time cell analysis and classical end-point assays were used. Flow cytometry was employed to determine the effect of different HIPEC drug solvents on tumor cell properties. Effectiveness of peritoneal perfusate patient samples and defined oxaliplatin-containing solutions proved limited but reproducible. HIPEC simulations for 30 min reduced the normalized cell index below 50% with peritoneal perfusates from merely 3 out of 9 patients within 72 h, indicating full-thickness cytotoxic effects. Instead, prolonging HIPEC to 1 h enhanced these effects and comprised 7 patients’ samples, while continuous drug exposure invariably resulted in complete cell death. Further, frequently used drug diluents caused approximately 25% cell size reduction within 30 min. Prolonging oxaliplatin exposure improved effectiveness of HIPEC to eliminate micrometastases in our preclinical model. Accordingly, insufficient penetration depth, short exposure time, and the physicochemical impact of drug solvents may constitute critical factors.The combined use of cytoreductive surgery (CRS) and hyperthermic intraperitoneal chemotherapy (HIPEC) for colorectal cancer (CRC) peritoneal metastasis has been supported by a leadoff randomized controlled trial (RCT) in 2003 showing prolonged overall survival for this treatment compared to palliative chemotherapy alone [1]. For decades this combined treatment approach has been in use for peritoneal metastases of different cancers [2]. The theoretical rationale of HIPEC is the elimination of residual tumor cells remaining after CRS through the locoregional administration of chemotherapy, which is justified by the assumption that a compartmental effect caused by the so-called peritoneal–plasma barrier prevents the penetration of intravenous drugs into the abdominal space [3]. The most frequent indication for this treatment has been peritoneal metastasis originating from ovarian cancer or CRC as well as pseudomyxoma peritonei. Accordingly, CRS and HIPEC has recently been shown to improve overall survival in a large cohort study in patients with pseudomyxoma peritonei [4] and can be considered the standard of care in this rare malignant syndrome [5]. In spite of long-standing practice, evidence from RCTs was lacking that could confirm survival benefits for adding HIPEC treatment to CRS alone. Only in 2018 an RCT demonstrated an increase in overall survival through adding 90-min cisplatin-based HIPEC to surgery in selected patients suffering from peritoneal metastasis that originated from platinum-sensitive ovarian cancers [6]. Of note, HIPEC treatment remained very heterogeneous over years [2,7]. In this context, the necessity of uniform and comparable treatment protocols has been rarely addressed, although this has been done for HIPEC with oxaliplatin (OX) in the treatment of CRC [8]. In contrast, attempts of standardization are commonplace for the surgical approach of CRS in peritoneal metastasis [9] and a required learning curve for improvement is well-established [10,11]. Amongst the numerous protocols in use, one of the most frequently applied drugs for peritoneal metastases in CRC is OX [7,12]. However, the recently published PRODIGE 7 study failed to show a survival benefit for the addition of 30-min OX-based HIPEC compared to surgery alone [13]. This multicenter RCT had used 460 mg OX (i.e., 230 µg/mL ≙ 579 µM OX) for open HIPEC (coliseum technique) and 360 mg OX (i.e., 180 µg/mL ≙ 453 µM OX) for closed HIPEC, diluted in dextrose solvent with each 2 L of perfusate filled into the abdomen. Since this outcome was unexpected, many discussions ensued as to why the HIPEC procedure performed so surprisingly poor [14,15,16]. Remarkably, basic and/or pharmacological research that may have shed more light in this context was unavailable. It should be acknowledged that CRS and HIPEC constitute a complex compound treatment with many variables, hence the single contribution of each of the components remains poorly dissected [17].For this reason, we were interested to investigate how different OX-containing solutions would perform under simulated HIPEC conditions in a ~100 µm thick micrometastasis model. Already in previous work we had established an impedance-based real-time cell analysis (RTCA) assay, enabling continuous monitoring of cancer cells. Hereby, a thorough characterization of patient perfusate samples taken during HIPEC procedures was performed [18]. At this time, we had used respective OX-containing samples diluted with 50% serum-containing cell culture medium and left them incubated with OAW42 ovarian cancer cells. This assay unanimously showed an elimination of the exposed cell layer within 72 h. The aim of the present study was now to assess HIPEC treatment in a preclinical model to separately characterize its effects within the multimodal therapy approach. To this end, we incubated a ~100 µm thick layer of platinum-sensitive tumor cells at 42 °C for 30 min or 60 min with well-defined sample materials previously obtained during HIPEC treatment of patients. We also used OX at clinically relevant concentrations diluted in different solvents, conforming with established HIPEC protocols. This also encompassed the exact OX dosages and diluents that have been reported for the PRODIGE 7 trial [13].Studies on materials sampled during patient treatment were approved by the institutional review board at Tübingen University (project number: 367/2013BO1). All patients gave their written informed consent before study inclusion. Patient characteristics and treatment details can be assessed elsewhere [18]. Drug solvent circulated through the abdomen of patients was collected prior to addition of OX (0 min) as well as OX-containing samples taken at 5, 10, 15, 20, 25, and 30 min after starting the HIPEC procedure. For patient 8 and 9, samples were only obtained in 10-min intervals. Samples were stored at −80 °C until usage. Cellular debris and other impurities were cleared by centrifugation (5 min at 13,000× g) before use.Materials and methods were used as established before [18], with slight modifications. As described previously, platinum-sensitive OAW42 cells (European Collection of Authenticated Cell Cultures, Salisbury, UK) were grown under appropriate conditions, seeded and used as a model system to monitor effects of OX in real-time [18]. In brief, cells were grown in cell culture medium (DMEM; Dulbecco’s Modified Eagle Medium, high glucose; Gibco/Life Technologies, Carlsbad, CA, USA, adding 10% fetal calf serum (FCS); Sigma-Aldrich/Merck Life Science, St. Louis, MO, USA; 100 U/mL penicillin G; PAA, Pasching, Austria; and 100 µg/mL streptomycin; PAA). Cell cultures were periodically tested for mycoplasma using commercially available polymerase chain reaction (PCR) kits (Minerva Biolabs, Berlin, Germany). An RTCA device (xCELLigence SP; Roche, Grenzach-Wyhlen, Germany) was employed as previously described [18]. After a calibration of the device with 100 µL DMEM (blank values), 5 × 104 OAW42 cells/ well were seeded into dedicated 96-well plates (E-plate 96; ACEA, San Diego, CA, USA) and left to adhere for 24 h. Subsequently, fluid was discarded from wells and adherent cells were washed once with warmed Dulbecco’s phosphate-buffered saline (PBS; Gibco Life Technologies, Carlsbad, CA, USA). Then, either 200 µL of peritoneal perfusate obtained during HIPEC from patients or defined concentrations of OX (Oxaliplatin-GRY/ oxaliplatin 5 mg/mL; Teva, Petach Tikwa, Israel/ Fresenius Kabi, Bad Homburg, Germany; dose range: 5.6–230 µg/mL ≙ 14–579 µM OX) diluted in peritoneal dialysis fluid (PDS; Physioneal 40 Glucose 2.27% m/v; Baxter, Deerfield, IL, USA) or dextrose 5% in water (D5W; Glucosteril 5%; Fresenius Kabi) were added. Likewise, adequate controls were employed, including 1% (v/v) Triton X-100 (Sigma) as a positive control (lysis control for dead cells) and negative controls with either PDS or DMEM only. HIPEC conditions were subsequently simulated by incubating cells for 30 min or 60 min at 42 °C in an ambient air incubation shaker (Infors, Bottmingen, Switzerland) with slight movement (50 rotations per min; rpm). Afterwards, liquids were discarded by flicking and cells were washed twice in PBS and then cultivated in appropriate medium. All samples were analyzed at least in duplicates and impedance was measured continuously in 15-min intervals. Technical errors and outliers were removed. For inter-experiment comparability, the cell index was set to 1 immediately before simulated HIPEC treatment (normalized cell index; nCI). Results were analyzed using RTCA software (V. 1.2.1). GraphPad prism software (V. 7.01; GraphPad software Inc., La Jolla, CA, USA) was used for presentation of results. In addition, biorender software was used for visualizations (www.biorender.com). Findings of multiple experiments were combined, when adequate.To confirm effects detected by RTCA on cell viability, CTB Cell Viability Assay (Promega, Mannheim, Germany) was performed. OAW42 and HT29 cells were seeded in 24-well plates (3.15 × 105 cells) in a volume of 500 µL in triplicates and incubated overnight at 37 °C and 5% CO2 in a humidified atmosphere. To mimic different HIPEC treatment conditions, DMEM was discarded, cells were washed with PBS and then incubated for 30 min at 42 °C under ambient air conditions with OX (at concentrations of 45, 90, 180, and 230 µg/mL; ≙ 113, 227, 453, and 579 µM OX) diluted either in D5W, PDS, or DMEM (control). Following treatment, solutions were discarded and replaced by 500 µL of fresh DMEM after washing with PBS and incubated for 72 h (37 °C; 5% CO2). In parallel, OX-spiked DMEM remained on the cells to verify OX toxicity on OAW42 and HT29 cells after 72 h continuous exposure (37 °C; 5% CO2). As positive controls, cell death was induced by lysing cells with 1% (v/v) Triton X-100 (Roth, Karlsruhe, Germany) for 10 min immediately prior to CTB staining. To determine cell viability, 100 µL of assay reagent were added and gently mixed. Fluorescence measurement was performed 1 h after incubation (37 °C; 5% CO2) for each cell line with the Synergy HT microtiter plate reader (BioTek Instruments Inc., Winooski, VT, USA; record fluorescence; excitation wavelength: 530/25 and emission wavelength: 590/35, adjusted sensitivity: 35). CTB assays were repeated in 3 independent experiments.OAW42 and HT29 cells were seeded in 24-well plates (3.15 × 105 cells/ well) in a volume of 500 µL as triplicates and incubated overnight (37 °C; 5% CO2). To mimic HIPEC treatment, medium was discarded, and cells were washed with PBS and then incubated for 30 min at 42 °C under ambient air conditions with OX diluted either in D5W, PDS, or DMEM (see OX concentrations given above). After treatment, solutions were discarded and replaced by 500 µL of fresh DMEM after washing with PBS and incubated for 72 h (37 °C; 5% CO2). In a parallel experiment, OX-spiked DMEM remained on the cells to verify OX effects on OAW42 and HT29 cells after 72 h exposure (37 °C; 5% CO2). As positive control, cell death was induced by lysing cells with 1% (v/v) Triton X-100 (Roth, Karlsruhe, Germany) for 10 min immediately prior to SRB staining. Finally, growth inhibition was evaluated by SRB assay. In brief, medium was discarded, and each well was washed once with ice-cold PBS and fixed with 10% trichloroacetic acid (TCA) for 30 min at 4 °C. After washing with tap water, cells were dried at 40 °C overnight. Then proteins were stained for 10 min with SRB reagent (0.4% (w/v) in 1% (v/v) acetic acid; CAS 3520-42-1, Sigma-Aldrich) and after removing unbound dye with tap water followed by 1% (v/v) acetic acid, dried again at 40 °C. Protein-bound dye was resolved with 10 mM Tris base (pH 10.5). After 10 min incubation at room temperature, optical density was measured in triplicates (80 µL volume/ well) in 96-well plates with a Synergy HT microtiter plate reader (BioTek Instruments; measurement wavelength 550 nm, reference wavelength 620 nm). Data represent the mean of optical density values related to DMEM treated control cells. SRB assays were repeated in 3 independent experiments.In order to obtain a serial dilution, OAW42 cells were seeded at different numbers of 12.5 × 103, 20 × 103, 25 × 103, 35 × 103, and 50 × 103 cells/ well (96-well flat bottom plate) and incubated in 200 µL DMEM. After 24 h of cell culture, cells were carefully washed with PBS and subsequently treated with fixation buffer (BioLegend, San Diego, CA, USA) for 10 min at room temperature, carefully washed again and kept at 4 °C prior to measurements. Experiments were performed twice. The thickness of cell layers was obtained measuring z-stacks on a Nikon Ti Eclipse microscope (by an unbiased observer) using 10× magnification. The analysis was performed with the NIS-Elements (Nikon, Tokyo, Japan) or ImageJ software V. 1.52 h. OAW42 cells were seeded in 96-well plates (1 × 105 cells/ well) in either 200 µL of DMEM, PDS, or D5W and kept for 30 min or 60 min at 42 °C. Immediately prior to flow cytometric analysis, 7-amino-actinomycin D viability staining solution (7-AAD; BioLegend) was added to each well at a final concentration of 600 ng/mL. Dead cells (7-AAD-positive) as well as doublets were excluded. Forward scatter area (FSC-A) served as an indicator for cell size. Samples were analyzed using a FACSCanto II (BD Biosciences, Heidelberg, Germany) and data analysis was performed using FlowJo_V9 software (FlowJo LCC, Ashland, OR, USA). Flow cytometric analysis of cell size under HIPEC conditions was repeated in independent experiments.The 2-dimensional (2D) micrometastasis model as established here aims to recreate the clinical conditions prevailing during HIPEC in a laboratory setting. Not only are cells exposed to hyperthermia and realistic drug dosages but also exposure time and respective solvents can be modelled with either patient samples or OX-containing solutions prepared appropriately in this preclinical model. Of note, the used RTCA system allows continuous monitoring of cell elimination and addresses the clinically relevant question of effectiveness in peritoneal micrometastases after optimal CRS (Figure 1). For this reason, our model system used a cell layer with defined thickness, requiring a penetration depth of about 100 µm for measurable effects (Figure S1).The RTCA readout is generated by measuring electron flow between an array of electrodes located at the bottom of each well in a specific 96-well E-Plate (Figure 1a). An intact cell layer isolates the electrodes and thereby impairs electron flow. Accordingly, impedance increases, which is indicated by a rising cell index (Figure 1b,c). This effect is proportional to the cell number and morphology of cells covering the electrodes at the well bottom (Figure 1e) [19]. In our model, relevant measurable effects require full thickness defects of the cell layer (decline in cell index Figure 1d). An nCI value of 0.5 specifies that the initially prevailing impedance measured at normalization has been bisected.First, we used sample materials that were obtained during OX-based HIPEC from 9 patients and that have been extensively characterized previously [18]. As shown before, the label-free RTCA assay enables monitoring cells continuously with good temporal resolution. However, now we intended to model HIPEC conditions more precisely and exposed ovarian cancer cells (OAW42) to respective aliquots from HIPEC patient samples for 30 min or 60 min at 42 °C (50 rpm shaking) to simulate respective conditions, subsequently washing cells and extending cell culture with fresh culture medium and regularly assessing nCI values (Figure 2a).Previously published experiments have shown that these exact OX-containing samples invariably eliminated the seeded cells, when continuously exposing them for 72 h to HIPEC perfusates, even when diluting them through supplementation of 50% medium [18].In our modified HIPEC model, most aliquots from the previously characterized patient samples [18] proved insufficient to effectively eliminate the cancer cells in our micrometastasis model within 72 h, when HIPEC was simulated for 30 min. Strikingly, in 6 out of 9 investigated patients, none of the obtained OX-containing samples tested was able to reduce impedance below 50% of initial values (nCI < 0.5) within 3 days following HIPEC simulation (Figure 2b). Prolonging exposure time to 1 h considerably improved cytotoxic effects but did not prove generally effective, lowering nCI < 0.5 when using samples from 7 out of 9 patients (Figure 2c). Observed effects were also stronger in those samples that were taken earlier after HIPEC initiation during clinical patient treatment. Further, using the overall means from different patient samples and comparing the different conditions to each other (Figure 2d) revealed clear differences between 30 min and 60 min of OX-based HIPEC simulation. When considering effects after 30 min exposure, both controls and HIPEC perfusates sampled at different time points showed comparable mean nCI readings ranging from 0.84 to 0.96 at 72 h following the treatment. In contrast, prolonging the procedure to 60 min resulted in mean nCI readings of 0.33 with OX-containing HIPEC perfusates sampled in patients after 10 min as well as decreasing mean nCI values to 0.61 (20 min) and 0.57 (30 min) for respective samples obtained later. Although it has to be mentioned that the source data partly show a bimodal distribution as well as a large variance (see also Figure 2c), this evaluation supports improved effects with increased HIPEC duration as well as underscoring patient individual differences.The differences in our micrometastasis model for the perfusates sampled in patients at several different time points during HIPEC and used for short-term (30 min) and prolonged (60 min) HIPEC simulations are striking. Looking at the RTCA results from perfusate samples in 2 exemplary patients, we can discern in patient 3 (Figure 3a) that with none of the OX-containing perfusates the exposed cell layer could be effectively eliminated following 30 min of simulated HIPEC, as shown by persistently elevated impedance readings. Prolonging the treatment duration to 1 h could rescue these effects and showed a decrease below nCI = 0.5 for the samples obtained during the first 15 min of clinical HIPEC treatment in this patient, whereas the other samples (20–30 min) proved ineffective. In contrast, patient 1 already showed a respective decrease of nCI to baseline values after 30 min of simulated HIPEC for all samples obtained during the first 20 min of the HIPEC procedure in the clinic (Figure 3b). Here, a prolonged exposure of cells to HIPEC samples over 1 h showed impedance decreasing to baseline values in all OX-containing perfusates tested.For a comparison of aliquots from the same sample materials of the 2 presented patients, contrasting both 30 min and 60 min of simulated HIPEC to continuous exposure, we annotated the time points when nCI = 0.5 was reached in an overview chart (Figure 3c; patient samples are aliquots from the materials previously characterized and data of continuous incubation were previously published by Löffler et al. [18]).On a side note, the perfusate used for continuous exposure of cells had to be diluted with medium (50%), since this is inevitably required to sustain cell culture. Respective samples therefore contain only half of the OX concentrations compared to those samples used for 30 min and 60 min of simulated HIPEC (Figure 3c). Still, continuous exposure to OX-containing HIPEC perfusates unanimously led to nCI decreases below 0.5. This was usually reached within 48 h with a slight waning effect over time for perfusates sampled later during HIPEC in patients. A complete dataset with results from the other patients is provided in the Supplementary Materials (Figures S2–S15). Further, it has been established that only 0.05 μM OX is sufficient to kill 50% of directly exposed OAW42 cells (LC50) within 72 h [20], whereas usual OX concentrations used during HIPEC are above 90 µM and reach up to 579 µM according to the PRODIGE 7 protocol [13].Next, we prepared defined OX-containing solutions that comprised accurate amounts of OX in different diluents, aiming to recreate the clinical conditions prevailing during HIPEC as authentically as possible. To this end, we used a concentration range of OX, encompassing the final drug dosages according to most established HIPEC protocols [16], explicitly including those used in the PRODIGE 7 study [13] and conforming with previously published HIPEC models [21].We assessed the effects of respective OX concentrations diluted either in D5W or PDS in our 100 µm thickness micrometastasis model after simulating the HIPEC procedure for 30 min. Subsequently, drugs were removed, and cells were washed and cultured in medium for the long-term.Here, we did not observe any substantial treatment induced decrease in nCI after 30 min of OX-based HIPEC using either D5W (Figure 4a) or PDS (Figure 4b) as a solvent within 96 h following treatment, with the exception of the highest dosage of 230 µg/mL OX diluted in PDS. Prolongation to 60 min simulated HIPEC showed effective for 180 µg/mL and 230 µg/mL OX when diluted with PDS (see Supplementary Materials: Figure S16). However, 60 min HIPEC exposure with OX diluted in D5W proved technically unfeasible due to recurring detachment of the treated cell layer.Again, when exposing cells continuously to OX at the specified concentrations diluted in cell culture medium (here doubled concentrations were used to account for 50% dilution with medium), this intervention proved effective to reduce nCI below 0.5 for a broad concentration range. For example, concentrations of 11.2–230 µg/mL OX proved effective when diluted in D5W and for 22.5–230 µg/mL in PDS. For the dose range of 22.5–45 µg/mL OX it appears that dilution in PDS reached nCI = 0.5 slightly faster than with D5W. Further higher drug concentrations seemed associated with reaching nCI = 0.5 earlier than when using a lower drug concentration. However, measurable effects still took several hours to appear.To confirm the RTCA results previously obtained, we performed 2 different well-established conventional end-point assays. The fluorometric assay CTB is based on the conversion of resazurin to resorufin, occurring only in living cells [22], and a cytotoxicity assay using SRB, which binds stoichiometrically to proteins under mild acidic conditions [23]. Cells were densely seeded to reach comparable conditions as used previously in the RTCA assays. To broaden our analysis, we also added the human colon cancer cell line HT29 [24], which is part of the NCI60 human tumour cell line anticancer drug screen panel and therefore well characterized [25]. The LC50 for OX in HT29 has been established at 72.44 µM [26]. After simulated HIPEC for 30 min with 45–230 µg/mL OX diluted in PDS and subsequently culturing the washed cells in medium, both the OAW42 and the HT29 cell lines showed no reduction of mean cell viability below 50% (LC50) in none of the assessed conditions within 3 days after treatment (Figure 5a). This finding applied to both assays performed. Repeating the experiments under identical conditions with OX diluted in D5W solution failed also to induce LC50 at 72 h following treatment for all tested concentrations (Figure 5b). In contrast, when OX was spiked into cell culture medium (to allow for extended cultivation) and remained with the cells, both cell lines showed clearly enhanced induction of cell death (Figure 5c). Cell viability reproducibly fell below LC50 values within 3 days after exposure to all drug concentrations investigated, with the exception of 45 µg/mL OX in medium when tested on HT29 cells. Findings proved reproducible as the results were comparable for both assays.OAW42 cells were exposed to D5W and PDS at 42 °C for either 30 min or 60 min and cell size was measured by flow cytometry, since respective solvents constitute the most frequently used drug diluents for HIPEC with OX. Significant cell size reduction of about 25% (in FSC-A) was observed after 30 min and 60 min of incubation at hyperthermic conditions with both solvents tested (D5W and PDS), when compared to cells maintained in cell culture medium (Figure 6). The observed effects are comparable between both solvents and showed only a slightly more increased shrinkage after 1 h as compared to 30 min exposure.Over the past decades, survival in metastatic CRC has widely improved. This is also true for metastases limited to the peritoneum, which were considered unresponsive to chemotherapy and therefore a generally palliative disease stage in the past [27]. Here CRS with HIPEC has relevantly enhanced patient prognosis. However, in CRC this complex compound treatment is currently highly controversial, since the randomized controlled PRODIGE 7 trial could not establish survival benefits for patients through adding OX-based HIPEC following CRS [13], whilst increasing morbidity and duration of hospital stay. Further, HIPEC using OX was unsuccessful in an another RCT in the adjuvant setting, where no benefit could be shown for HIPEC to decrease the incidence of metachronous peritoneal metastasis following primary CRC resection [28]. Certainly, it should not go unmentioned that CRS alone was able to increase median overall survival of patients to 41 months according to the PRODIGE 7 trial, which is unprecedented for this disease [13]. These unexpectedly high survival rates in both study cohorts have raised questions pertaining to the study design and the included patient population as well as to tumor biology. These issues include a high rate of crossovers into the HIPEC group and the inclusion of highly selected patients with heavy pretreatment (including systemic OX administration), therefore potentially weakening the validity of these trial results and limiting conclusions that can be drawn regarding HIPEC ineffectiveness [29,30]. In any case, HIPEC with OX currently faces pressure for justification.Besides the prevailing controversies concerning the clinical use of HIPEC and the interpretation of the PRODIGE 7 trial results, including patient selection and speculations on acquired resistance of peritoneal metastases against OX [14,15,16,31], it should be acknowledged that preclinical research has never been a priority in the implementation of this treatment modality. The consequence of this is an abundance of different HIPEC protocols in clinical use [2,7] and lacking evidence regarding the precise mode of action of HIPEC. In accordance, taking a step back and systematically investigating the effects of the varying treatment parameters of HIPEC in a preclinical setting before implementation in clinical trials is a sensible strategy and should serve as an example [32]. One central but hitherto poorly answered question is the tissue penetration depth of drugs used for HIPEC and more importantly their respective biological effects on cells [17]. Low penetration depth has been identified early on as a potential limiting factor of intraperitoneal drug delivery [33], which was assumed to be a few millimeters at most [34]. While only sparse clinical data are available in this context and in spite of research that suggests an increased penetration depth with heat for OX [35], recent findings suggest ineffectiveness of 30 min OX-based HIPEC in organoids [21]. Therefore, to better assess tissue penetration depth during HIPEC and respective cytotoxic effects in a dedicated 2D model, we repurposed a RTCA assay [18], allowing the continuous assessment of a ~100 µm thick cell layer by using the platinum sensitive cell line OAW42 [20]. For our model we simulated HIPEC conditions by incubating cells with either peritoneal perfusate samples obtained during HIPEC treatment in patients or OX-containing solutions prepared at clinically relevant dosages by dilution in peritoneal dialysis or dextrose solutions [7,13]. Hence, we assume this model can elucidate the direct cytotoxic effects of HIPEC on peritoneal micrometastases. Moreover, it should be mentioned that OX cytotoxicity is well-established for the used cell lines, even at a fraction of the concentrations employed in our experiments [20,36].The observed results underscore that short-term HIPEC with OX exerts only very slight effects in a model system that requires drugs to penetrate a distance of about 1/10 of a millimeter and to affect exposed cells in this way. A respective lack of effectiveness was witnessed in platinum-sensitive cells after 30 min HIPEC simulation with OX diluted in dextrose solutions over a wide concentration range, also encompassing the drug amounts employed in the PRODIGE 7 trial [13]. According to these results, OX only proved effective when diluted in peritoneal dialysis solution at the highest concentration tested and OX diluted in dextrose solution generally failed to decrease impedance values relevantly, thus suggesting a lack of a profound impact on residual metastatic tissue. Using sample materials obtained during HIPEC from 9 patients and characterized to contain a calculated mean of 93.7 μg/mL OX [18], no relevant effects were observed with materials from 6 patients following 30 min of HIPEC simulation within a measurement window of 72 h. In contrast, prolonging exposure to 1 h relevantly improved cytotoxic effects and proved effective with materials from 7 patients. Further assessment of overall means from different patient samples emphasized lacking effectiveness of 30 min OX-based HIPEC in simulations as well as substantial improvements after prolongation to 60 min, although with large variance underscoring relevant individual differences of the sample materials assessed. Recently a study in patient-derived CRC organoids likewise showed that prolonged exposure to OX at lower concentrations was more cytotoxic than short-term exposure at higher concentrations [37]. This notion generally conforms with our results, since the effects observed after simulated HIPEC with materials from 9 patients could be improved by prolonging simulated HIPEC treatment to 60 min. Regarding the RTCA assay with OX-containing solutions prepared at clinically relevant dosages, unfortunately we cannot conclude comprehensively on HIPEC prolongation to 1 h, since with dextrose solution our model proved unsuitable due to cell detachment. Nevertheless, the results from short-term exposure were supported by the assessment of 2 different cell lines in 2 well-established end-point assays using a comparable setting: following 30 min of OX-based HIPEC simulation, less than 25% reduction of cell viability was observed compared to values below 50% cell viability when continuously exposed to OX even at low concentrations. Likewise, long term exposure to OX even at relatively low concentrations unanimously eliminated the exposed cell layer in our RTCA model.Evidently, as HIPEC involves a variety of factors, these may become relevant in the context of different drugs used and influence molecular mechanisms relevant for anti-cancer effects. Importantly, both the exposure period and heat have been implicated in HIPEC effects. For instance, in vitro studies with OX have shown increased drug uptake and DNA damage with heat, resulting in increased apoptosis but still required 60 min exposure for effectiveness in respective CRC models, whereas, e.g., mitomycin C lacked such synergistic effects [38]. Further, it is known that heat inhibits DNA repair, e.g., through inhibiting poly(ADP-ribose)-polymerase (PARP1), and therefore it can sensitize cells for chemotherapy [39,40]. Another important determinant in HIPEC is the solvent used to dilute the drugs. Here we observed significant cell size reduction of about 25% after 30 min, persisting also at 60 min under hyperthermic conditions with both tested solvents (PDS and D5W as compared to medium). We speculate that the observed changes are most likely caused by a fluid shift to the extracellular space, which also occurs during HIPEC treatment and may impair drug penetration into tissues.As a limitation of this modelling study it should be acknowledged that this system is artificial, since merely effects on platinum-sensitive cancer cells are assessed and the peritoneal microenvironment cannot be imitated. In reality, increased interstitial fluid pressure in the tumor [41], tissue cohesion, edema formation, and other factors may likely complicate the picture. Using sample materials obtained during HIPEC from patients also has immanent limitations, since, e.g., we found previously that the administered OX does react with contents of peritoneal dialysis solutions by forming new compounds [18], which may affect cellular uptake and molecular drug mechanisms.Apart from this, the introduced model has several advantages compared to alternative approaches. These include the possibility to investigate HIPEC effects continuously in the 2D model under specific consideration of the penetration depth, which is in contrast to 3D models that show considerably higher degrees of freedom, e.g., with regard to organoid size and other properties [21,37]. Due to the biosensors´ functional assessment a spatial resolution is ensured, informing only about the cell viability of those cells located at the well bottom. Thereby effects on cells can be directly investigated, avoiding the extrapolation of biological effects from drug penetration depth, which itself comes with immanent limitations and idiosyncrasies [42]. Our assay is robust and shows consistency and good reproducibility, allowing for systematic comparisons to be performed. Further continuous measurements provide a good temporal resolution and are therefore more sensitive to changes occurring with delay and superior to end-point assays. Hitherto, there are few good preclinical models for laboratory use and our model is unique in several aspects, adding a new tool to HIPEC research.Overall, preclinical models for HIPEC are scarce and mainly restricted to animal models so far [43,44], whereas adequate model systems have only been established very recently [21,32,37]. If the findings observed for OX hold true for other tumor entities and drugs used for HIPEC and under normothermic conditions remains unclear, which is why our model system may be a worthwhile addition for future research.Overall, our investigations showed severely limited effects of short-term OX-based HIPEC, in line with previous research. We now show that the desired effects could be improved by prolonged exposure. Strikingly, our simulations of HIPEC treatment in platinum-sensitive cell lines using excessive OX dosages proved insufficient to eradicate even minimal tumor cell accumulations. Thus, the presented work may support the notion that short-term OX-based HIPEC is ineffective to eliminate peritoneal micrometastasis and may hint to possible causes for this failure. Our findings suggest that drug distribution into deeper tissue layers may be a crucial factor as well as exposure time but also the solvents used for OX dilution in HIPEC may substantially alter drug induced effects.Based on the restricted penetration depth of most HIPEC drugs [45], systematic preclinical research is warranted. Taken together, HIPEC—in contrast to CRS—remains an experimental approach for treatment of peritoneal metastasis originating from CRC and new evidence is urgently needed.The following are available online at https://www.mdpi.com/article/10.3390/cancers14051158/s1, Figure S1: Assessment of the thickness of an OAW42 cell layer seeded at different densities; Figure S2: RTCA: Pat. 2: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S3: RTCA: Pat. 2: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S4: RTCA: Pat. 4: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S5: RTCA: Pat. 4: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S6: RTCA: Pat. 5: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S7: RTCA: Pat. 5: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S8: RTCA: Pat. 6: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S9: RTCA: Pat. 6: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S10: RTCA: Pat. 7: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S11: RTCA: Pat. 7: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S12: RTCA: Pat. 8: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S13: RTCA: Pat. 8: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S14: RTCA: Pat. 9: Exposure of OAW42 cells to OCS for 30 min at 42 °C; Figure S15: RTCA: Pat. 9: Exposure of OAW42 cells to OCS for 60 min at 42 °C; Figure S16: RTCA: Oxaliplatin (OX)-spiked into PDS 60 min at 42 °C; Figure S17: RTCA: Continuous exposure of OAW42 cells to OCS in PDS; Figure S18: RTCA: Continuous exposure of OAW42 cells to OCS in D5W.Conceptualization: T.J., S.V., S.B. and M.W.L. Data curation: N.S., M.B. Formal analysis: N.S., M.B., A.T., F.H., J.K., S.V. and M.W.L. Investigation: N.S., M.B., B.O., F.H., J.K. and M.W.L. Methodology: N.S., M.B., S.B. and M.W.L. Validation: N.S., C.Y., M.B., B.O., J.K., M.Q. and M.W.L. Visualization: N.S., C.Y., M.B., and M.W.L. Project administration: C.Y., A.T., T.J., K.T., S.V. and M.W.L. Funding acquisition: I.K., A.K. and M.W.L. Resources: I.K., K.T., H.-G.R., M.S. and A.K. Supervision: C.Y., A.T., K.T., H.-G.R., S.V., M.S., A.K. and S.B. Writing—original draft: N.S., C.Y. and M.W.L. Writing—review and editing: M.B., B.O., A.T., F.H., J.K., T.J., I.K., K.T., M.Q., H.-G.R., S.V., M.S., A.K. and S.B. All authors have read and agreed to the published version of the manuscript.We acknowledge support by Open Access Publishing Fund of University of Tübingen. This work was supported by the Robert Bosch Stiftung (Stuttgart, Germany), and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) under Germany’s Excellence Strategy—EXC 2180—390900677. This work was supported by grants from the Deutsche Forschungsgemeinschaft (DFG, SFB 685) to I. Königsrainer and A. Königsrainer. The funders had no role in the design of the study, in the collection, analyses, or interpretation of data, in the writing of the manuscript, or in the decision to publish the results.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of the Medical Faculty at the University of Tübingen and of the University Hospital (project number: 367/2013BO1).Informed consent was obtained from all patients involved in the study.The data presented in this study is contained within the article or Supplementary Materials further information is available on request from the corresponding author. The authors would like to thank Jan Franko, Des Moines, Iowa for insights, encouragement, and helpful discussions. We also thank Jürgen Winter, Department of General, Visceral and Transplant Surgery, University Hospital of Tübingen, for excellent technical support as well as Jürgen Weinreich, and Philipp Horvath, for support with obtaining sample materials. The authors would like to acknowledge the contributions of Sebastian P. Haen, University Medical Center Hamburg-Eppendorf, to this study and we would like to commemorate him and his scientific excellence, exceptional personality and friendship.S. Venturelli and M. Burkard were supported by a grant from the Else-Übelmesser-Stiftung (D.30.21947; reference: GzV 1.14) of the University of Tübingen unrelated to the present work. M. W. Löffler and A. Königsrainer have received a research grant by RanD S.r.l., a manufacturer of devices and consumables for HIPEC, unrelated to the present work. The other authors declare no potential conflict of interest.Background on the used micrometastasis model and readout of the used real-time impedance-based cell analysis (RTCA) assay. (a) The experiment in a 96-well E-plate is started 24 h before the planned HIPEC treatment. Blank values with only cell culture medium are measured. Under these conditions, electrons can flow freely between the gold electrodes located at the well bottom (baseline; low impedance). (b) Subsequently, cells are added to each well and left to attach for 24 h forming a ~100 µm thick cell layer (impedance increases to reach a plateau, since electron flow is heavily impaired through the isolation effects of the added cells). At this point, before HIPEC simulation (b), impedance is normalized to 1 (nCI = 1) to allow for comparability between independent experiments. (c) HIPEC simulation is performed by incubating cells with OX-containing solutions for 30 min or 60 min at 42 °C with slight movement. Afterwards samples are removed, cells washed and supplemented with fresh medium. Measurement is continued for 4 days. (d) If the complete full cell layer is affected, and cells become penetrable in full thickness, while only debris remains, electrons can flow freely, and impedance decreases to baseline values. (e) If cells persist and isolation effects impairing free electron flow remain, impedance diminishes depending on intact cells left. An nCI = 0.5 specifies the value when the initial impedance has been bisected and can be determined as a function of time. The thickness of the seeded cells was assessed and the linearity of thickness according to cell numbers seeded per well established (Figure S1). For a relatable scale, the edge length of a grain of salt is about 300 µm.HIPEC simulations using patient samples (RTCA assay). (a) Exemplary impedance readings in 12-h intervals. The displayed time point 72 h after HIPEC simulation is marked by a red border. (b,c) Simulated HIPEC (42 °C at 50 rpm shaking) was performed, either for 30 min (b) or for 60 min (c) with 5 × 104 OAW42 cells per well. (d) Means of replicate values shown in (b,c) are annotated as dots and respective overall mean values of different patient materials are shown as a bar plot and annotated (left bars/ blue colors: 30 min exposure in HIPEC simulations; right bars/ green colors: 60 min exposure in HIPEC simulations). Due to the large variance and partly bimodal data distribution, any statistical significance testing was omitted and means chosen as a measure of central tendency. X-axis: positive control (+ ctrl.) with Triton X-100, negative control (− ctrl.) (light grey coloration) and HIPEC solutions obtained from patients (Pat.) 1–9 each: samples before/pre (dark grey coloration), 10 min (dark blue/ dark green coloration), 20 min (medium blue/ medium green coloration), and 30 min (light blue/ light green coloration) after adding OX to the HIPEC circuit during patient treatment. Y-axis: nCI determined at 72 h since beginning of measurements after HIPEC treatment. Each colored arrow marks an average decrease in nCI (72 h) below 0.5 (black horizontal line). Depiction of mean values with standard deviation, number of replicates: 2–5.HIPEC simulations for 30 min or 60 min using patient samples (RTCA assay) (a,b) Simulated HIPEC (42 °C at 50 rpm shaking) was performed for 30 min (upper graph) and 60 min (lower graph) with 5 × 104 OAW42 cells per well. Positive control (+ ctrl.) with Triton X-100, negative control (− ctrl.) with PDS (peritoneal dialysis solution), MEM (cell culture medium), and HIPEC solutions obtained from patient (Pat.) 3 (a) and Pat. 1 (b). Samples: before (pre), 5, 10, 15, 20, 25, and 30 min after adding OX to the HIPEC circuit during clinical patient treatment. Y-axis: nCI determined until 96 h after restarting measurements following HIPEC treatment. (c) Duration until nCI = 0.5 was reached after 30 min or 60 min simulated HIPEC or by continuous incubation (nota bene: dilution with 50% MEM) with respective samples. Depiction of mean values with standard deviation, number of replicates: 2–4. Respective RTCA readings with HIPEC sample materials obtained from Pat. 2 and Pat. 4–9 are provided as Supplementary Materials (Figures S2–S15).HIPEC simulation for 30 min using prepared OX-containing solutions (RTCA assay) (a,b) Simulated HIPEC (42 °C and 50 rpm shaking) was performed for 30 min with 5 × 104 OAW42 cells per well. Positive control (+ ctrl.) with Triton X-100, negative control (− ctrl.) and prepared solutions with specified OX concentrations in dextrose 5% (D5W) (a) or peritoneal dialysis solutions (PDS) (b). Y-axis: nCI determined until 96 h (h) after restarting measurements following HIPEC treatment. (c) Duration until nCI = 0.5 was reached after 30 min simulated HIPEC or when continuously incubated (dilution with 50% cell culture medium (MEM)) with respective samples. Depiction of mean values with standard deviation, number of replicates: 2–3. Respective RTCA readings with continuous OX exposure are provided as Supplementary Materials (Figures S17 and S18).HIPEC simulation for 30 min using prepared OX-containing solutions diluted either in PDS or D5W (CTB and SRB assay). OAW42 cells (red coloration) as well as HT29 cells (blue coloration) were used at a density of 3.15 × 105 cells per well (in a 24-well format) to recreate the conditions encountered in RTCA assays before. HIPEC was simulated for 30 min at 42 °C with prepared solutions containing the specified amounts of OX diluted either in PDS (a) or in D5W (b). After exposure, cells were washed and subsequently cultured in cell culture medium (MEM) for another 3 days. Further, respective cells were incubated continuously with the specified amounts of OX, spiked into MEM to allow for continuous cell culture and heated likewise (30 min at 42 °C) followed by 72 h cell culture (c). Thereafter, the CTB cell viability assay (left graphs) or the SRB cytotoxicity assay (right graphs) were used. Cells were normalized to cells treated identically with D5W, PDS, and MEM only (viab. ctrl.). Positive control (+ ctrl.) was carried out with 1% (v/v) Triton X-100. Statistical analysis was performed using the Dunnet’s multiple comparison test, confidence interval 95%. ns: p ≥ 0.05; **: p < 0.01; ***: p < 0.001 vs. the respective viability control. The LC50 threshold is marked with a black line. Depiction of mean values with standard deviation from 3 independent experiments, with triplicate values assessed in each experiment performed.Flow cytometry of OAW42 cells after exposure to different drug diluents for 30 min or 60 min under hyperthermic conditions. OAW42 cells were incubated with D5W (blue coloration) or with PDS (green coloration) at 42 °C and slight shaking (30 rpm) for 30 min (a) or 60 min (b) and compared to untreated control cells cultured in medium (MEM; grey coloration). Histograms from flow cytometry showing cell counts versus forward scatter area (FSC-A) (left panels) and a comparison of FSC-A between the respective solvents (right panels). Significant differences are marked by an asterisk (*: p < 0.05; Bonferroni corrected Student’s t-test). Each data point represents the mean value of 3 replicates in independent experiments.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01159.txt

@@ -0,0 +1,5 @@
+These authors contribute equally to this work.In this pilot study, we aimed to investigate the use of deep learning for the classification of whole-slide images of liquid-based cytology specimens into neoplastic and non-neoplastic. To do so, we used a large training and test sets. Overall, the model achieved good classification performance in classifying whole-slide images, demonstrating the promising potential use of such models for aiding the screening processes for cervical cancer.Liquid-based cytology (LBC) for cervical cancer screening is now more common than the conventional smears, which when digitised from glass slides into whole-slide images (WSIs), opens up the possibility of artificial intelligence (AI)-based automated image analysis. Since conventional screening processes by cytoscreeners and cytopathologists using microscopes is limited in terms of human resources, it is important to develop new computational techniques that can automatically and rapidly diagnose a large amount of specimens without delay, which would be of great benefit for clinical laboratories and hospitals. The goal of this study was to investigate the use of a deep learning model for the classification of WSIs of LBC specimens into neoplastic and non-neoplastic. To do so, we used a dataset of 1605 cervical WSIs. We evaluated the model on three test sets with a combined total of 1468 WSIs, achieving ROC AUCs for WSI diagnosis in the range of 0.89–0.96, demonstrating the promising potential use of such models for aiding screening processes.According to the Global Cancer Statistics 2020 [1], cervical cancer is the fourth leading cause of cancer death in women, with an estimated 342,000 deaths worldwide in 2020. However, incidence and mortality rates have declined over the past few decades due to either increasing average socioeconomic levels or a diminishing risk of persistent infection with high risk human papillomavirus (HPV) [1]. In developed countries, cervical cytology screening systems have been organised to reduce mortality from cervical cancer [2,3,4,5,6,7,8,9].The introduction of cervical cancer screening led to a fall in associated mortality rates; however, there is some evidence that the conventional smear method for screening is not consistent in reliably detecting cervical intraepithelial neoplasia (CIN) [10,11,12]. This is because conventional cervical smears, when spread on glass slides, tend to have the cells of interest mixed with blood, debris, and exudate. A number of new technologies and procedures are becoming available in various screening programs (e.g., liquid-based cytology (LBC), automated screening devices, computer-assisted microscopy, digital colposcopy with automated image analysis, HPV testing). The LBC technique preserves the cells of interest in a liquid medium and removes most of the debris, blood, and exudate either by filtering or density gradient centrifugation. The other advantages in LBC are the availability of residual material for HPV and other molecular tests and the connection with automated screening devices. ThinPrep (Hologic, Inc., Marlborough, MA, USA) and SurePath (Becton Dickinson, Inc., Franklin Lakes, NJ, USA) for LBC specimen preparation have been approved by the US Food and Drug Administration (FDA), and it has also been adopted by the cervical screening programme in the UK. Moreover, the ThinPrep collection vial has been approved by the FDA for direct testing for HPV, which is particularly useful for managing women whose Pap smear tests show atypical squamous cells (ASCs) [4,13].In 1998, the FDA approved the FocalPoint Slide Profiler (Becton Dickinson, Inc.) as a primary automated screener for cervical smears, followed by approval in 2002 for use with SurePath slides. In 2003, the FDA approved the ThinPrep Imaging System (Hologic, Inc.) as a primary screener for ThinPrep Pap slides. The FocalPoint uses algorithms to measure cellular features (e.g., nuclear size, integrated optical density, nuclear to cytoplasmic ratio, and nuclear contour) for the diagnosis of squamous and glandular lesions [14]. In the US, the American Society of Cytopathology (ASC) established guidelines for automated Pap test screening using the ThinPrep Imaging System and the FocalPoint GS Imaging System [15]. However, there are some issues with the current automated screening support systems. A multi-institutional feasibility study in Japan validated the usefulness of FocalPoint for cervical cytology automated screening quality control and showed that it was useful for NILM (Negative for Intraepithelial Lesion or Malignancy) cases, but on the other hand, 2174 (18.1%) of 12,000 specimens were judged to be unmeasurable and were not evaluated [16]. In the US, unmeasured rates were reported to be as low as 2.5% [17], 5.9% [18], and 4.8% [19], while in Brazil, the unmeasured rate was very high at 30.8% [20]. In order to use FocalPoint, it was reported that the unmeasured ratio can be suppressed to a low value by adjusting a specimen preparation method(s) including staining [16]. However, in routine clinical practice, there are many screening facilities that do not (or cannot) stain specimens accordingly to adjust for FocalPoint, as reported in Japan and Brazil [16,20].The sensitivity of conventional cytology cervical cancer screening for detecting pre-invasive squamous and glandular lesions (pre-invasive intraepithelial lesions) is clearly far from perfect. It has been reported that most studies of the conventional Pap test were severely biased, and it was only moderately accurate and did not achieve concurrently high sensitivity and specificity (i.e., sensitivity ranged from 30% to 87% and specificity ranged from 86% to 100%) [21]. Moreover, the sensitivity of conventional cervical cytology is less than ideal for invasive cancers, with a wide range (45% to 76%), and false-negative or false-unsatisfactory rate in conventional smears was 50% [22]. These studies indicate that many women with cervical cancer have a history of one or more negative cervical cytology reports. As a background of these results, the interobserver reproducibility of cervical cytology is less than perfect. The reproducibility of 4948 monolayer cytologic interpretations was moderate (kappa = 0.46; 95% confidence interval (CI), 0.44–0.48) among four categories of diagnosis (i.e., negative, ASC-US, LSIL, and over HSIL) by multiple well-trained observers [23]. In the same study, the greatest disagreement in monolayer cervical cytology involved ASC-US interpretations. Of the 1473 original interpretations of ASC-US, the second reviewer concurred in only 43.0% [23].Whole-slide images (WSIs) are digitisations of the conventional glass slides obtained via specialised scanning devices (WSI scanners), and they are considered to be comparable to microscopy for primary diagnosis [24]. A routine scanning of LBC slides in a single layer of WSIs would be suitable for further high throughput analysis (e.g., automated image based cytological screening and medical image analysis) [25]. The advent of WSIs led to the application of medical image analysis techniques, machine learning, and deep learning techniques for aiding pathologists in inspecting WSIs. Deep-learning-based applications ranged from tasks, such as cancer diagnosis from WSIs, cell classification, and segmentation of nuclei, to patient stratification and outcome prediction [26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44]. For cytology, in particular, only recently have there been investigations for applying deep learning on large datasets of cervical WSIs Holmström et al. [45], Lin et al. [46], Cheng et al. [47].In this pilot study, we trained a deep learning model, based on convolutional and recurrent neural networks, using a dataset of 1605 cervical WSIs. We evaluated the model on three test sets with a combined total of 1468 WSIs, achieving ROC AUCs for WSI diagnosis in the range of 0.89–0.96.This is a retrospective study. A total of 3121 LBC ThinPrep Pap test (Hologic, Inc.) conventionally prepared cytopathological slide glass specimens of human cervical cytology were collected from a private clinical laboratory in Japan after cytopathological review of those specimens by cytoscreeners and pathologists. The cases were selected mostly at random so as to reflect a real clinical scenario as much as possible; we have also collected cases so as to compile a test set with an equal balance of neoplastic and NILM. The cytoscreeners and pathologists excluded cases that had poor scanned quality (n=32). Each WSI diagnosis was observed by at least two cytoscreeners and pathologists, with the final checking and verification performed by a senior cytoscreener or pathologist. All WSIs were scanned at a magnification of ×20 using the same Aperio AT2 digital whole-slide scanner (Leica Microsystems, Osaka, Japan) and were saved in SVS file format with JPEG2000 compression.Table 1 breaks down the distribution of the dataset into training, validation, and test sets. The split was carried out randomly taking into account the proportion of each label in the dataset. A clinical laboratory that provided LBC cases was anonymised. The test sets were composed of WSIs of full agreement, clinical balance, and equal balance LBC specimens. The full agreement test set consisted of NILM and neoplastic LBC cases whose obtained diagnoses were fully agreed by two independent cytoscreeners in different institutes. The clinical balance test set consisted of 95% NILM and 5% neoplastic LBC cases based on a real clinical setting [48,49]. The equal balance test set consisted of 50% NILM and 50% neoplastic LBC cases. NILM and neoplastic LBC cases for clinical and equal balance test sets were collected based on the diagnoses provided by the clinical laboratory. The cases in the clinical and equal balance test sets were only based on the diagnostic reports. From these two test sets, we have also created their reviewed counterparts (clinical balance reviewed and equal balance reviewed), where two independent cytoscreeners viewed all the cases and the ones they had a disagreement on were removed (see Table 1).Senior cytoscreeners and pathologists who perform routine cytopathological screening and diagnoses in general hospitals and clinical laboratories in Japan manually annotated 352 neoplastic WSIs from the training sets. Coarse annotations were obtained by free-hand drawing. (Figure 1 using an in-house online tool developed by customising the open-source OpenSeadragon tool at https://openseadragon.github.io/ (accessed on 10 January 2020), which is a web-based viewer for zoomable images.) On average, the cytoscreeneers and pathologists annotated 150 cells (or cellular clusters) per WSI.Neoplastic WSIs consisted of ASC (atypical squamous cell), LSIL (low-grade squamous intraepithelial lesion), HSIL (high-grade squamous intraepithelial lesion), CIS (carcinoma in situ), ADC (adenocarcinoma), and SCC (squamous cell carcinoma), except for the NILM. For example, on the HSIL (Figure 1A–D) and SCC (Figure 1E–H) WSIs, cytoscreeners and pathologists performed annotations around the neoplastic cells (Figure 1B–D,F–H) based on the representative neoplastic epithelial cell morphology (e.g., increased nuclear/cytoplasmic ratio, abnormalities of nuclear shape, hyperchromatism, irregular chromatin distribution, and prominent nucleolus). On the other hand, the cytoscreeners and pathologists did not annotate areas where it was difficult to cytologically determine that the cells were neoplastic. The NILM subset of the training and validation sets (1301 WSIs) was not annotated and the entire cell spreading areas within the WSIs were used.The average annotation time per WSI was about an hour. Annotations performed by the cytoscreeners and pathologists were modified (if necessary), confirmed, and verified by a senior cytoscreener.Our deep learning models consisted of a convolutional neural network (CNN) and a recurrent neural network (RNN) that were trained simultaneously end to end. For the CNN, we have used the EfficientNetB0 architecture [50] with a modified input size of 1024 × 1024 px to allow a larger view; this is based on cytologists’ input that they usually need to view the neighbouring cells around a given cell in order to diagnose more accurately. We then performed 7 × 7 max pooling with a stride of 5 × 5. The output of the CNN was reshaped and provided as input to an RNN with a gated recurrent unit Cho et al. [51] model of size 128, followed by a fully connected layer. We used the partial fine-tuning approach [52] for the tuning the CNN component, where only the affine weights of the batch normalisation layers are updated while the rest of the weights in the CNN remain frozen. We used the pre-trained weights from ImageNet as starting weights. Figure 2 shows a simplified overview of the model. The RNN component was initialised with random weights.WSIs tend to contain a large white background that is not relevant for the model. We therefore start the preprocessing by eliminating the white background using Otsu’s method [53] applied to the greyscale version of the WSIs.For training and inference, we then proceeded by extracting 1024 × 1024 px tiles from the tissue regions. We performed the extraction in real-time using the OpenSlide library [54]. To perform inference on a WSI, we used a sliding window approach with a fixed-size stride of 512 × 512 px (half the tile size). This results in a grid-like output of predictions on all areas that contained cells, which then allowed us to visualise the prediction as a heatmap of probabilities that we can directly superimpose on top of the WSI. Each tile had a probability of being neoplastic; to obtain a single probability that is representative of the WSI, we computed the maximum probability from all the tiles.During training, we maintained an equal balance of positively and negatively labelled tiles in the training batch. To do so, for the positive tiles, we extracted them randomly from the annotated regions of neoplastic WSIs, such that within the 1024 × 1024 px, at least one annotated cell was visible anywhere inside the tile. For the negative tiles, we extracted them randomly anywhere from the tissue regions of NILM WSIs. We then interleaved the positive and negative tiles to construct an equally balanced batch that was then fed as input to the CNN. In addition, to reduce the number of false positives, given the large size of the WSIs, we performed a hard mining of tiles, whereby at the end of each epoch, we performed full sliding window inference on all the NILM WSIs in order to adjust the random sampling probability such that false positively predicted tiles of NILM were more likely to be sampled.During training, we performed real-time augmentation of the extracted tiles using variations of brightness, saturation, and contrast. We trained the model using the Adam optimisation algorithm [55], with the binary cross entropy loss, beta1=0.9, beta2=0.999, and a learning rate of 0.001. We applied a learning rate decay of 0.95 every 2 epochs. We used early stopping by tracking the performance of the model on a validation set, and training was stopped automatically when there was no further improvement on the validation loss for 10 epochs. The model with the lowest validation loss was chosen as the final model.For the interobserver concordance study, a total of 10 WSIs (8 NILM cases and 2 neoplastic cases) of cervical LBC already reported by a clinical laboratory were retrieved from the records. Using the in-house on-line web virtual slide application, a total of 16 cytoscreeners (8 have over 10 years experiences and 8 have less than 10 years experiences) have reviewed the 10 WSIs and reported in subclasses (NILM, ASC-US, ASC-H, LSIL, HSIL, SCC, ADC).The deep learning models were implemented and trained using the open-source TensorFlow library [56].To assess the cytopathological diagnostic concordance of cytoscreeners, we performed the Fleiss’ kappa statistic, which is a measure of inter-rater agreement of a categorical variable [57] between two or more raters. We calculated the kappa values using Microsoft Excel 2016 MSO (16.0.13029.20232) 64 bit. The scale for interpretation is as follows: ≤0.0, poor agreement; 0.01–0.20, slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; 0.81–1.00, almost perfect agreement. AUCs were calculated in python using the scikit-learn package [58] and plotted using matplotlib [59]. The 95% CIs of the AUCs were estimated using the bootstrap method [60] with 1000 iterations.The true positive rate (TPR) was computed as
+        (1)TPR=TPTP+FN
+        and the false positive rate (FPR) was computed as
+        (2)FPR=FPFP+TN
+        where TP, FP, and TN represent true positive, false positive, and true negative, respectively. The ROC curve was computed by varying the probability threshold from 0.0 to 1.0 and computing both the TPR and FPR at the given threshold.We adapted the training code from https://github.com/tensorflow/models/tree/master/official/vision/image_classification (accessed on 14 February 2020).The aim of this retrospective study was to train a deep learning model for the classification of neoplastic cervical WSIs. We trained a model that consists of a convolutional and a recurrent neural network using a dataset of 1503 WSIs for training and 150 for validation. We evaluated the model on three test sets with a combined total of 1468 WSIs. Figure 3 shows the resulting ROC curves, and Table 2 lists the resulting ROC AUC and log loss, as well as the accuracy, sensitivity, and specificity computed at a probability threshold of 0.5. Table 3 shows the confusion matrix. The model achieved a good performance overall, with ROC AUCs of 0.96 (0.92–0.99) on the full agreement, 0.89 (0.81–0.96) on the clinical balance reviewed, and 0.92 (0.89–0.94) on the equal balance reviewed test sets.Our deep learning model satisfactorily predicted neoplastic epithelial cells (Figure 4C–G) in cervical LBC (Figure 4A,B) specimen. The heatmap image shows true positive predictions (Figure 4B–D) of neoplastic epithelial cells. In contrast, in low probability tiles (Figure 4H,I), two independent cytoscreeners confirmed there were no neoplastic epithelial cells.Our model satisfactorily predicted NILM cases (Figure 5A,B) in cerevical LBC specimen. The heatmap image shows true negative predictions (Figure 5B,D,E) of neoplastic epithelial cells. In both zero (Figure 5C) and very low probability tiles (Figure 5D,E), there are no neoplastic epithelial cells.A cytopathologically diagnosed NILM case (Figure 6A) was false positively predicted for neoplastic epithelial cells (Figure 6B). The heatmap image (Figure 6B) shows false positive predictions of neoplastic epithelial cells (Figure 6C,E) with high probabilities. Cytopathologically, there are parabasal cells with a high nuclear cytoplasmic (N/C) ratio (Figure 6C,D) and cell clusters of squamous epithelial cells with cervical gland cells with high N/C ratios (Figure 6E), which could be a major cause of false positive.To evaluate the practical interobserver variability among cytoscreeners, we have asked a total of 16 cytoscreeners (8 are over 10 years experiences and 8 are less than 10 years experiences) to review the same 10 LBC WSIs, which consist of 8 NILM and 2 neoplastic cases already diagnosed by a clinical laboratory. The results of each cytoscreener were summarised in Table 4. The Fleiss’ kappa statistics were summarised in Table 5. There was poor to moderate concordance in assessing subclass, with Fleiss’ kappas of NILM (range: 0.042–0.755), neoplastic (range: 0.098–0.500), and all cases (range: 0.364–0.716). On the other hand, there was poorly to almost perfect concordance in assessing binary class, with Fleiss’ kappas of NILM (range: 0.073–0.815), neoplastic (1.000), and all cases (range: 0.568–0.861). Interestingly, there was a robust higher concordance in both subclass and binary class among cytoscreeners over 10-year experiences. However, overall, there was poor concordance in assessing NILM cases (range: 0.042–0.073).In this pilot study, we trained a deep learning model for the classification of neoplastic cells in WSIs of LBC specimens. The model achieved overall a good performance, with ROC AUCs of 0.96 (0.92–0.99) on the full agreement, 0.89 (0.81–0.96) on the clinical balance reviewed, and 0.92 (0.89–0.94) on the equal balance reviewed test sets.Looking at the interobserver concordance among cytoscreeners in Table 4, it is obvious that there is considerable interobserver variability, with the poor concordance in NILM cases even for binary classification (NILM vs. neoplastic). In addition, there is the problem of human fatigue due to the continuous observation of a large number of cases. Therefore, when considering future accuracy control, it may be necessary to conduct screening using deep learning model(s) with guaranteed accuracy, such as the results of this study, at least in the binary classification (NILM vs. neoplastic), and to conduct detailed assessments by cytoscreeners and cytopathologists in the subclassification (e.g., NILM, ASC-US, ASC-H, LSIL, HSIL, SCC, and ADC).From our results in Figure 2, it was obvious that there was interobserver variability among cytoscreeners in different clinical laboratories and hospitals. Clinical balance and equal balance test sets were prepared based on diagnostic (screening) reports from a clinical laboratory. The only difference between clinical balance and clinical balance-reviewed (same as equal balance and equal balance-reviewed) was whether it was additionally reviewed by two more cytoscreeners in different clinical laboratories and hospitals or not. All scores (ROC-AUC, accuracy, sensitivity, and specificity) were increased in clinical balance-reviewed and equal balance-reviewed test sets as compared to clinical balance and equal balance test sets (Figure 2). Hence, our deep learning model would be helpful for standardising in the screening process.In routine cervical cancer screening at clinical laboratories and hospitals, it is difficult to introduce a screening programme dependent on cervical smears due to poor human cytoscreener resources. LBC techniques opened new possibilities for a systemic cervical cancer screening. LBC slides are amenable to high throughput automated analysis. Especially for the detection of rare events on LBC slides, WSI and subsequent image analysis is of crucial importance for guaranteeing a standardised high-quality read out [25]. Practical automated cervical cytology screening devices have been under development since the 1950s. The technological development in semi-automated screening devices for cervical cancer screening is very rapid; however, currently, no machines are available to provide a fully automated screening by computer without human intervention. There are two FDA-approved semi-automated slide scanning devices on the market; these systems are the BD FocalPoint GS Imaging System and the HOLOGIC ThinPrep Imaging System. Both are designed to perform computer-assisted analysis of cellular images followed by location-guided screening of limited fields of view. FocalPoint-assisted smear reading has been proposed prior to conventional manual reading; the latter may be unnecessary for cases reported as No Further Review (NFR) and would be required for cases reported as Review (REV) [61]. FocalPoint-assisted practice showed statistically superior sensitivity and specificity when compared to conventional manual smear screening for the detection of HSIL and LSIL [14,62,63]. However, ASC-US sensitivity and specificity were not significantly different between FocalPoint-assisted practice and conventional screening [62]. Overall, in neoplastic slides (ASC-US, LSIL, and HSIL) by FocalPoint-assisted practice, sensitivity was in the range of 81.1–86.1% and specificity was in the range of 84.5–95.1% [62]. The other study showed that FocalPoint-assisted reading was comparable to conventional reading, and the very low observed negative predictive value of an NFR report (0.02%) suggested that these cases might safely return to periodic screening [61]. The ThinPrep Imaging System (TIS) is an automated system that uses location-guided screening to assist cytoscreeners in reviewing a ThinPrep Pap LBC slides [64]. TIS scans the LBC slides and identifies 22 fields of view (FOVs) on each slide based on optical density measurements and other features [64]. It has been reported that TIS was ideally suited to the rapid screening of negative cases; however, the sensitivity and specificity of the TIS (85.19% and 96.67%, respectively) were equivalent to those of manual screening (89.38% and 98.42%, respectively) [65]. In another study, for diagnostic categories of neoplastic slides (ASC-US, LSIL, and HSIL) by TIS practice, sensitivity was in the range of 79.2–82.0% and specificity was in the range of 97.8–99.6% [64].As shown in Figure 2, our LBC cervical cancer screening deep learning model exhibited around 90% accuracy (in the range of 89–91%), 86% sensitivity (in the range of 84–89%), and 91% specificity (in the range of 90–92%) in full agreement, clinical balance-reviewed, and equal balance-reviewed test sets; those scores were as well or better than the existing assistance systems mentioned above.In the present study, we have trained a deep learning model for the classification of neoplastic cervical LBC in WSIs. We have evaluated the model on three test sets achieving ROC-AUCs for WSI diagnosis in the range of 0.89–0.96. The main advantage of our deep learning model is that the model can be used to evaluate the cervical LBC at the WSI level. Therefore, our model is able to infer whether the cervical LBC WSI is NILM (non-neoplastic) (Figure 5) or neoplastic (Figure 4). This makes it possible to use a deep learning model such as ours as a tool to aid in the cervical screening process, which could potentially be used to rank the cases by order of priority. After which the cytoscreeners will need to perform full screening and subclassification (e.g., ASC-US, ASC-H, LSIL, HSIL, SCC, ADC) on neoplastic output cases after the primary screening by our deep learning model, which could reduce their working time as the model would have highlighted the potential suspected neoplastic regions, and they would not have to perform an exhaustive search through the entire WSI.F.K. and M.T. contributed equally to this study; F.K., S.I. and M.T. designed the studies; F.K., N.H, T.I., A.F., S.I. and M.T. performed experiments and analysed the data; N.H., T.I., A.F. and S.I. performed cytopathological diagnoses and reviewed cases; F.K. and M.T. performed computational studies; F.K., S.I. and M.T. wrote the manuscript; M.T. supervised the project. All authors reviewed and approved the final manuscript.The authors received no financial supports for the research, authorship, and publication of this study.The experimental protocol in this study was approved by the ethical board of the private clinical laboratory. All research activities complied with all relevant ethical regulations and were performed in accordance with relevant guidelines and regulations in the clinical laboratory. Due to the confidentiality agreement with the private clinical laboratory, the name of the clinical laboratory cannot be disclosed.Informed consent to use cytopathological samples (liquid-based cytology glass slides) and cytopathological reports for research purposes had previously been obtained from all patients and the opportunity for refusal to participate in research had been guaranteed by an opt-out manner.The datasets used in this study are not publicly available due to specific institutional requirements governing privacy protection; however, they are available from the corresponding author and from the private clinical laboratory in Japan on reasonable request. Restrictions apply based on the data use agreement, which was made according to the Ethical Guidelines for Medical and Health Research Involving Human Subjects as set by the Japanese Ministry of Health, Labour, and Welfare.We thank cytoscreeners and pathologists who have been engaged in reviewing cases, annotations, and cytopathological discussion for this study.F.K. and M.T. are employees of Medmain Inc. The authors declare no conflict of interest.Representative manually drawing annotation images for neoplastic labels on liquid-based cytology (LBC) slides. The LBC case (A) was diagnosed as HSIL (high-grade squamous intraepithelial lesion) based on the representative neoplastic squamous epithelial cells with increase in nuclear/cytoplasmic ratio and nuclear atypia (B–D). The LBC case (E) was diagnosed as SCC (squamous cell carcinoma) based on the representative neoplastic squamous epithelial cells with HSIL features (F–H). Representative neoplastic cells were roughly annotated using in-house on-line drawing tools.Method overview. (a) Large 1024 × 1024 are extracted from the WSIs; for the neoplastic WSIs, tiles are extracted only from annotated regions, while from NILM WSIs, tiles are extracted randomly from any region. (b) The tiles are then used to create random balanced batches used to train the model, which is composed of a CNN and an RNN and are trained simultaneously. During inference, the model is applied on all of the tiles of the WSIs in a sliding window fashion, and the WSI label is predicted based on the maximum probability from all of the tiles.ROC curves for the three test sets.A representative example of neoplastic true positive prediction outputs on a liquid-based cytology (LBC) case from test sets. In the neoplastic whole-slide image (WSI) of LBC specimen (A), the heatmap image (B) shows a true positive prediction of neoplastic epithelial cells in high probability tiles (C,D), which correspond, respectively, to neoplastic epithelial cells (E–G) equivalent to HSIL (high-grade squamous intraepithelial lesion). On the other hand, in low probability tiles (H,I) of the same heatmap image (B), there are no evidence of neoplastic cells.A representative example of neoplastic true negative prediction outputs on a liquid-based cytology (LBC) case from test sets. In the NILM (negative for intraepithelial lesion or malignancy) whole slide image (WSI) of LBC specimen (A), the heatmap image (B) shows true negative prediction of neoplastic epithelial cells which correspond, respectively, to non-neoplastic epithelial cells (C). Moreover, in very low probability tiles (D,E) of the same heatmap image (B), there are no evidence of neoplastic cells.A representative example of neoplastic false positive prediction outputs on a liquid-based cytology (LBC) case from test sets. Cytopathologically, (A) is a NILM (negative for intraepithelial lesion or malignancy) whole-slide image (WSI) of LBC specimen. The heatmap image (B) exhibited false positive predictions of neoplastic tiles (C,E). In (C), there are parabasal cells with a slightly high nuclear cytoplasmic (N/C) ratio with dense chromatin appearance due to the cellular overlapping (D). In (E), there are cell clusters of squamous epithelial cells and cervical gland cells with slightly high N/C ratios and a dense chromatin appearance due to the cellular overlapping.Distribution of WSIs into training, test, and validation sets.ROC AUC, log loss, accuracy, sensitivity, and specificity results on the test sets.Confusion matrix.Cytopathological evaluations for 10 LBC WSIs by diagnostic report (Dx) and 16 cytoscreeners (CS) with their age and years of experience.Interobserver variability: kappa.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01160.txt

@@ -0,0 +1 @@
+Head and neck squamous cell carcinoma (HNSCC) is an aggressive and lethal disease. Despite diagnostic and therapeutic advances, the overall survival of patients with advanced HNSCC remains poor. Recently, microRNAs in extracellular vesicles (EV-miRNAs) have been proposed as essential regulatory molecules involved in HNSCC. EV-miRNAs may serve as disease biomarkers and represent a novel therapeutic target. This review summarizes the current understanding of the role of EV-miRNAs in HNSCC as well as their potential future clinical applications.MicroRNAs (miRNAs) are a class of small non-coding RNA molecules that play a pivotal regulatory role in a broad variety of biological processes. Dysregulation of miRNAs is associated with several human diseases, particularly cancer. Extracellular vesicles (EVs) are crucial components in intercellular communication. As part of the cargo of EVs, miRNAs are involved in EV-mediated cell-to-cell interactions, including promotion or suppression of tumor development. The knowledge on the molecular mechanisms and clinical importance of EV-miRNAs in head and neck squamous cell carcinoma (HNSCC) has rapidly grown over the past years. In the present review, the current understanding regarding the effect of EV-miRNAs on HNSCC tumorigenesis is summarized, which includes effects on tumor proliferation, angiogenesis, invasion and metastasis, the tumor microenvironment, immune modulation, and treatment resistance. EV-miRNA-based biomarkers in liquid biopsies such as blood and saliva may open up new possibilities for employing EV-miRNAs for screening and early diagnostics as well as disease monitoring. Future perspectives include the promise of EV-miRNAs as a novel therapeutic target.Head and neck squamous cell carcinoma (HNSCC) is the 8th most common malignancy worldwide, with 790,000 patients diagnosed and 400,000 patients dying from this disease each year [1]. Classical risk factors for the development of HNSCC are nicotine and alcohol abuse. High risk serotype human papillomavirus (HPV) infection is an additional risk factor for HNSCC of the oropharynx [2,3]. The prevalence of HPV-positive oropharyngeal squamous cell carcinoma (OPSCC) is considerably higher in the USA and Western countries compared to low- and middle-income countries [4]. HPV-positive OPSCC is a distinct disease entity compared to HPV-negative OPSCC, with a favorable response to treatment and better overall survival (OS) (3-year OS 82.4% versus 57.1%, respectively) [5]. Currently, testing for HPV status in OPSCC is the only prognostic molecular test used in the clinical management of HNSCC [6].Despite attempts to improve the treatment outcome of HNSCC, the 5-year OS rate has remained unchanged over the last decade. Two-thirds of patients with HNSCC present with locoregional disease which signifies that the disease has metastasized to regional cervical lymph nodes [7,8]. Earlier stages of HNSCC are treated by surgery or radiotherapy alone, resulting in a 5-year OS of 70–90%. For locally advanced disease, multimodality treatment (a combination of surgery and/or (chemo)radiotherapy) is required with a 5-year survival rate lagging behind at 40–60% [9]. Whilst the recurrence rate for early-stage disease is 10–12%, around half of the patients with locally advanced disease experience disease recurrence within the first two years, either locoregionally or as distant metastases. Patients with recurrent and metastatic (R/M) HNSCC have a poor prognosis with a median OS of less than one year [10,11]. A more comprehensive understanding of the molecular mechanisms driving and characterizing HNSCC is urgently needed in order to improve disease outcome by better molecular diagnostics and more effective therapies.MicroRNAs (miRNAs) are a class of small non-coding RNAs, on average 22 nucleotides in length, that play a pivotal role in the post-transcriptional regulation of gene expression. MiRNAs can induce mRNA degradation and suppression of protein translation primarily through complementary base pairing with the 3′ untranslated region of their target mRNA. Aberrant expression of miRNAs has been associated with numerous human diseases, including cancer [12].Extracellular vesicles (EVs) are a heterogeneous population of phospholipid bilayer-enclosed particles that are released by most cell types and are widely distributed in the blood, saliva, and other body fluids [13]. EVs can be classified as exosomes or microvesicles according to their intracellular origin. The nano-sized exosomes are generated within endosomal compartments. Microvesicles bud directly from the plasma membrane [14]. EVs harbor a cargo consisting of proteins, mRNA, non-coding (nc) RNAs such as microRNAs and long nc-RNAs, DNA, and lipids. This cargo can be transferred to recipient cells, which constitutes an important form of physiological cell-to-cell communication [15]. Adversely, EVs are also involved in the development and progression of many diseases. Plasma of patients with cancer, including HNSCC, is known to be enriched in exosomes [16,17]. Over the past decade, it became clear that EVs participate in tumor progression by mediating the crosstalk between tumor cells and between stromal cells in the tumor microenvironment (TME) [14]. Notably, the EV-mediated transfer of miRNAs was shown to play a crucial role in advancing tumorigenesis by promoting angiogenesis, metastasis formation [18,19,20], TME reprogramming [21], immune tolerance [22,23], and therapy resistance [24]. By dissecting the tumor-secreted EV-miRNA profile, the importance of EV-miRNAs in cancer development is gradually being revealed. In this review, we summarize the current knowledge on EV-miRNAs in HNSCC. We also describe the potential applications of EV-miRNAs in novel treatment approaches and EV-miRNAs as diagnostic and prognostic biomarkers in HNSCC.MicroRNAs (miRNAs) are a class of small non-coding RNA molecules that play pivotal regulatory roles in numerous biological processes. Dysregulation of miRNAs is associated with several human diseases, particularly cancer. Research on the role of EV-miRNAs in cancer has unveiled a broad array of mechanisms by which EV-miRNAs are implicated in carcinogenesis. Pathogenic EV-miRNAs can be actively exported by parent cells and imported by destination cells as part of vesicle trafficking and intercellular communication [25]. An ongoing debate is what determines the miRNA content of EVs. As the relative miRNA composition in EVs is different from their parent cells, an active sorting mechanism into these vesicles is suggested [25,26]. It is proposed that AGO2 and other RNA-binding proteins such as hnRNPA2B and Y-box protein 1 are involved in the regulation of miRNA loading into EVs [27,28,29]. However, further investigations, particularly in vivo experiments, are still required for more definitive conclusions [30]. EV-miRNAs can act as oncomiRs, i.e., miRNAs of which overexpression is associated with the development of cancer, or as tumor suppressor miRNAs which are generally underexpressed in cancer. EV-miRNAs can regulate cell proliferation, migration, epithelial–mesenchymal transition (EMT), tumor proliferation, angiogenesis, and metastasis formation among others. EV-miRNAs can also be employed to modulate the tumor microenvironment as well as the immune system [31]. With the ability of EV-microRNAs to regulate gene expression both locally and distantly, coinciding with the non-immunogenic character of EVs themselves, EVs may serve as a drug delivery platform for microRNA-based therapies. As EV-miRNA expression profiles are different between healthy subjects and cancer patients, they may also be used for novel diagnostic tests, including cancer screening, as well as for disease monitoring [32].The following section will be focusing on EV-miRNAs in the development and progression of HNSCC. An overview of published reports is provided in Table 1 and Figure 1.Several studies have evaluated the role of EV-miRNAs in cancer cell proliferation and metastasis [50,51,52,53]. Still, many aspects concerning the exact mechanism of EV-miRNA transfer and its effects on recipient cells are unclear. Predominantly, the mechanism is reported that EV-miRNAs from donor cells modify the phenotype of recipient cells by epigenetic regulation of gene transcription. For example, Melo et al. have described how exosomes derived from patients with breast cancer can alter the transcriptome of normal cells by stimulating cell proliferation and tumor formation [54].In HNSCC, miR-101-3p enriched exosomes derived from human bone marrow mesenchymal stem cells (hBMSCs) overexpressing miR-101-3p, were able to suppress oral cancer cell proliferation and tumor growth both in vitro and in vivo by targeting the Collagen Type X Alpha 1 Chain gene (COL10A1), resulting in downregulation of Collagen X expression [33]. A recent study demonstrated that oral squamous cell carcinoma (OSCC) derived-exosomes containing miR-130b-3p could promote angiogenesis in HUVEC cells through targeting the Phosphatase and Tensin Homolog (PTEN) tumor suppressor gene. The oncogenic effect of miR-130b-3p on tumor growth and blood vessel formation was confirmed in a tumor xenograft mouse model [34].Invasion and metastasis formation are important aspects in HNSCC progression. It is a complex process involving cell invasion, secretion of extracellular matrix metalloproteinases (MMPs), epithelial–mesenchymal transition (EMT), and suppression of anoikis [8]. EV-miRNAs may contribute to metastasis formation in HNSCC in several ways.First, transfer of EV-miRNAs from highly invasive tumor cells to less invasive tumor cells can induce a pro-metastatic phenotype in recipient cells. For example, release of exosomes containing miR-342–3p and miR-1246 by a highly metastatic human oral cancer cell line was found to induce cell motility and invasive abilities in poorly metastatic cells. It was hereby reported that miR-1246 directly suppressed expression of the tumor suppressor gene DENN/MADD Domain Containing 2D (DENND2D) [35]. Similarly, metastatic OSCC cells can release exosomes containing miR-200c-3p, which can bind the downstream targets chromodomain helicase DNA binding protein 9 (CHD9) and Werner syndrome RecQ like helicase (WRN), inducing an invasive phenotype in prior non-invasive OSCC cells [36]. Another study showed that OSCC-derived EVs containing miR-21-5p enhanced the metastatic phenotype of OSCC cell lines and transformed normal gingival fibroblasts (NGFs) into cancer-associated fibroblasts (CAF) [37].As a second mechanism, certain conditions of cellular stress, such as hypoxia, can alter the miRNA composition of EVs. It is known that a high level of hypoxia is associated with poor prognosis and resistance to radiotherapy in HNSCC [55,56,57,58,59]. It is thought that under hypoxic conditions, tumor cells regulate the EV-miRNA content to modulate the tumor microenvironment and promote angiogenesis and metastasis [60,61]. For example, exosomes from hypoxic OSCC contained higher levels of oncomiRNA-21 compared to normoxic OSCC cells, which was dependent on activation of hypoxia-inducible factor (HIF)-1α and HIF-2α. MiR-21-rich exosomes induced OSCC cell migration, invasion, and expression of mesenchymal markers (Snail and Vimentin) and reduced the expression of epithelial marker E-cadherin both in vitro and in vivo [38].Finally, tumor cells can interact with other cellular components of the tumor microenvironment via EV-miRNAs. Hsieh et al. have shown that Snail, the transcription factor regulating EMT, induces miR-21-enriched exosomes by direct transcriptional activation of the MIR21 gene. MiRNA-21 abundant exosomes promoted M2-like polarization of macrophages as well as suppression of M1-markers [39]. This shift in tumor-associated macrophage (TAM) phenotype was associated with angiogenesis and tumor growth [62]. In another study, cancer-associated fibroblast (CAF)-derived exosomes transferred miR-382-5p to OSCC cells which enhanced OSCC cell motility and invasiveness [40].In HNSCC tumors, the TME is a complex and diverse mix of tumor cells and stromal cells, including CAFs, endothelial cells, and immune cells [8]. Mesenchymal stromal cells (MSCs), major cell components in the TME, significantly influence the development and progression of cancer [63,64]. MSCs have been found to migrate into tumors and develop into tumor-associated MSCs and cancer-associated fibroblasts [65,66,67]. CAFs can greatly impact the progression of HNSCC as they can produce a wide range of growth factors (e.g., hepatocyte growth factor (HGF), epidermal growth factor (EGF), and vascular endothelial growth factor (VEGF)), cytokines (such as IL-6), matrix metalloproteinases (MMPs), and chemokines which can drive tumor cell growth, angiogenesis and immune suppression [68,69]. It has been demonstrated that EV-miRNAs originating from CAFs are essential regulators of HNSCC progression. Based on the research of Yao-Yin et al., it was found that OSCC cells gained a more aggressive phenotype after exposure to miR-34a-5p-devoid exosomes derived from CAFs. Additionally, the transfer of miR-34a-5p suppressed the proliferation and motility of OSCC cells by targeting the AXL gene (AXL receptor tyrosine kinase) which led to inhibition of the EMT-involved AKT/GSK-3β/β-catenin/Snail signaling cascade [41]. MiR-382-5p overexpression was detected in CAFs compared to fibroblasts from adjacent normal tissue. Although the exact mechanism remains unclear, miR-382-5p containing CAF-derived exosomes were suggested to be responsible for OSCC cell migration and invasion [40]. CAF-derived miR-196a-rich exosomes were proposed to play a key function in regulating HNSCC cell survival and proliferation. MiR-196a targets inhibitor of growth 5 (ING5) and cyclin-dependent kinase inhibitor 1B (CDKN1B), conferring cisplatin resistance to HNSCC cells [42]. Another study compared the differential miRNA profiles between exosomes from CAFs and normal fibroblasts (NFs). MiR-3188 was shown to be the most downregulated miRNA in CAF-derived exosomes. The loss of miR-3188 in CAF-derived exosomes increased proliferation and inhibited apoptosis in HNSCC cells by de-repressing B-cell lymphoma 2 (BCL2) expression both in vitro and in vivo. Exosomes rich in miR-3188 impaired tumor development in HNSCC xenografts [43]. It was shown that under hypoxic conditions, tumor cells can induce CAF-like differentiation of fibroblasts through the release of EV-miRNAs. The overexpression of miR-192 and miR-215 in hypoxic HNSCC-derived EVs was mediated by NF-κB and HIF-1α, respectively. EV-miR-192/215, when taken up by fibroblasts, resulted in downregulation of Caveolin-1 (CAV1) expression, a tumor suppressor gene that regulates the CAF-like differentiation of fibroblasts through inhibition of Transforming Growth Factor (TGF)-β/SMAD signaling. In turn, CAF-like differentiation mediates the progression of tumor cells through a positive feedback loop [44].HNSCC tumors are generally highly infiltrated by both tumor-infiltrating lymphocytes (TILs), e.g., B-cells, T-cells, and natural killer (NK) cells, and myeloid-lineage cells, e.g., macrophages, dendritic cells, neutrophils, and myeloid-derived suppressor cells (MDSCs). Prior studies have revealed evidence of immune cell dysfunction within the tumor microenvironment and in the peripheral blood of patients with advanced HNSCC [8]. The strong immunosuppressive effects of the TME allow tumors to evade immune surveillance. Consequently, many currently developed therapeutic strategies aim to restore the immune response, for instance, by treatment with immune checkpoint inhibitors [70,71]. The immunosuppressive milieu within the TME is mediated either directly by HNSCC tumor cells or indirectly via the stroma and chemokine-induced recruitment and polarization of immune cells such as MDSCs [72]. The role of EVs in immune modulation has been reported in various tumor types. However, there is limited data on EV-miRNA involvement in immune suppression in HNSCC. Momen-Heravi et al. reported that stimulation of an OSCC cell line with various doses of alcohol enhanced release of EVs containing oncogenic microRNAs such as miR-21 and miR-27. Subsequent exposure of monocytes with EVs from OSCC cells treated with alcohol resulted in activation of the NF-κB pathway and pro-tumorigenic reprogramming of monocytes [45]. The most abundant innate immune cells in the TME are tumor-associated macrophages (TAMs), which play a key role in tumor progression [73]. Macrophages derived from monocytes can be categorized into classically activated (M1) macrophages, which produce pro-inflammatory cytokines to eliminate pathogens, and alternatively activated (M2) macrophages, which secrete anti-inflammatory cytokines controlling tissue repair and immunosuppression [74]. It has been known that TAMs exhibiting a polarized M2 phenotype facilitate tumor growth and progression [75]. TAMs originate from monocytic precursors, which can differentiate and become activated in response to several stimuli released by tumor or stromal cells [76]. Snail-overexpressing HNSCC cells can, by secretion of miR-21-rich exosomes, promote M2-like polarization of tumor-associated macrophages by miR-21-mediated suppression of transcription of target genes such as programmed cell death protein 4 (PDCD4) and IL12A [39].In adult human peripheral blood, γδ T cells represent a minor lymphocyte cell population comprising between ~0.5% and 16% of the total of CD3+ cells. They are also found in the gut- and skin-associated lymphoid systems and in organized lymphoid tissues [77]. γδ T cells have been shown to exhibit direct cytotoxicity against malignant cells and possess antigen-presenting properties, making them attractive candidates for tumor immunotherapy [78]. In contrast, pro-tumoral activities of γδ T cells have also been reported in several cancer types [79,80]. The precise mechanisms underlying the dual role γδ T cells are still obscure and further research is needed. The role of EV-miRNAs regarding γδ T cell function and expansion has been reported in OSCC. MiR-21 expression was significantly increased in exosomes from hypoxic OSCC cells compared to normoxic OSCC cells. In a normoxic environment, OSCC-derived exosomes could activate γδ T-cell expansion and cytotoxicity in a heat shock protein (HSP)70-dependent but dendritic cell-independent manner. In contrast, in a hypoxic environment, OSCC-derived, miR-21-rich exosomes enhanced the suppressive function of MDSCs through miR-21 mediated downregulation of PTEN levels and upregulation PD-L1 expression, which subsequently led to γδ T cell exhaustion [46]. A different study discovered that miR-9 was more abundant in exosomes from HPV positive HNSCC compared to HPV negative HNSCC. Exosomal miR-9 induced M1 polarization in macrophages through downregulation of peroxisome proliferator-activated receptor δ (PPARδ). The classically activated (M1) macrophages contain a higher level of inducible nitric oxide synthase (iNOS) and reactive oxygen species (ROS), which subsequently enhanced the radiosensitivity of HNSCC cells [47].Cisplatin (CDDP)-based chemotherapy is an integral part of the treatment of advanced HNSCC. Cisplatin resistance, which can be intrinsic or acquired during treatment, is one of the most challenging issues in treating patients with HNSCC [81]. Understanding resistance mechanisms is essential in predicting treatment outcome and overcoming drug resistance with new therapeutic strategies. The mechanisms of cisplatin resistance are numerous, including a reduced cellular uptake of cisplatin, increased cellular efflux, enhanced DNA repair in response to cisplatin-induced DNA damage, and anti-apoptotic factors [82,83]. Growing evidence indicates that tumor-derived EV-miRNAs can confer a cisplatin resistant phenotype to recipient cells [84,85,86]. Exosomal miR-21 released by cisplatin-resistant OSCC cells can turn cisplatin-sensitive OSCC cells into cisplatin resistant, both in vitro and in vivo, by targeting tumor suppressor PTEN and programmed cell death 4 (PDCD4) [48]. The presence of cancer stem cells (CSCs) in HNSCC tumors contributes to the modulation of the TME, tumor progression, and resistance to treatment [87,88]. According to Chen et al., EVs produced by oral cancer stem cells (CSCs) may have a role in the development of cisplatin resistance. CSCs are a small subpopulation of cells within tumors with the ability of self-renewal, differentiation and tumor formation. The presence of CSCs was found in many cancer types and has been demonstrated as a driver of poor clinical outcome due to their contribution to chemotherapy resistance and metastasis [89]. CSC-derived EVs harboring miR-21-5p and other oncogenic signaling molecules, were shown to activate the PI3K/mTOR/STAT3 signaling pathway, leading to the CDDP resistance of differentiated OSCC cells. EVs released by CSCs were demonstrated to induce a CAF phenotype in normal gingival fibroblasts (NGFs), which subsequently induced a malignant phenotype in surrounding OSCC cells. In contrast, treatment with ovatodiolide (OV), a bioactive component of the Anisomeles indica plant, known for its anti-inflammatory properties, could reverse these effects. Treatment with OV resulted in a decrease in the oncogenic cargo of CSC-EVs, suppression of OSCC tumorigenesis, inhibition of NGF-CAF formation and normalization of the TME as well as restoration of cisplatin sensitivity [37]. These results suggest that disrupting EV-mediated communication between CSCs, tumor cells, and stroma might be utilized to overturn CDDP resistance in OSCC.Docetaxel (DTX) is an anticancer drug with anti-tumor activity in numerous solid tumor types, including oral squamous cell carcinoma [90]. A recent study investigated the role of EV-miRNAs in docetaxel chemoresistance. Downregulation of miR-200c was associated with resistance to DTX and resulted in increased migration and invasion and decreased apoptosis in tongue squamous cell carcinoma (TSCC) cells. Overexpression of miR-200c in normal tongue epithelial cells (NTECs) resulted in the release of miR-200c abundant EVs. Exposure of DTX-resistant cells to miR-200c-rich EVs led to increased sensitivity to DTX in both in vitro and in vivo experiments by targeting of the tubulin beta 3 class III (TUBB3) and protein phosphatase 2 scaffold subunit Abeta (PPP2R1B) genes [49].There is emerging evidence in various types of cancer that the TME can also induce chemoresistance through EV-miRNAs [91,92,93]. A study has shown that miR-196a containing exosomes are released from CAFs, enhancing proliferation, survival and conferring cisplatin resistance to HNSCC cells by targeting the CDKN1B and ING5 genes. Additionally, it was shown that the packaging of miR-196a into CAF-exosomes was mediated by RNA binding protein (RBP) heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1) [42].HPV-positive HNSCCs are more radiosensitive and display a significantly favorable clinical outcome over their HPV-negative counterparts. Although the underlying mechanisms are not fully understood, multiple factors are considered to contribute to radiosensitivity. The potential underlying mechanisms include DNA repair capacity, activation of tumor cell repopulation pathways and the oxygenation level in the tumor [94,95]. In addition, the immune cell population within the TME is thought to affect tumor radiosensitivity [96,97]. As mentioned in Section 3.3, the study by Tong et al. described miR-9-driven M1 macrophage polarization in HPV-positive HNSCC, resulting in enhanced tumor radiosensitivity [47]. The abovementioned findings indicate novel therapeutic avenues to overcome treatment resistance by intervening in the EV-miR balance in the tumor and TME.Modulation of EV-miRNA-mediated vesicle trafficking constitutes a novel therapeutic target in cancer. So far, studies have focused on extracellular vesicles as a drug delivery system for miRNAs. The clinical success of conventional drug delivery systems such as peptides, polymers, lipid microparticles, and nanoparticles has been limited. Challenges include reaching the target tissue, crossing of the blood–brain barrier, and the effective engagement of intracellular targets. Furthermore, issues regarding toxicity and immunogenicity of non-natural delivery systems remain [98,99]. EVs have attracted tremendous attention in the context of biomolecule delivery platforms due to their ideal carrier system properties. The double-layered membrane of EVs protects its cargo from degradation and prolongs their circulation half-life [100]. Synthetic naked miRNAs have a short circulation half-life that could be increased by packaging into EVs, which offers protection from ribonucleases [101,102]. Another advantage is that EVs are able to traverse complex biological barriers such as the blood–brain barrier [102,103,104]. Importantly, when using autologous exosomes, minimal immunogenicity is exhibited, unlike viral gene transfer vectors or liposomes [102,105]. Using EV-miRNAs as a therapeutic approach has been investigated in various types of cancer [106,107,108]. The number of studies on HNSCC is, however, still limited (see Table 2 and Figure 2). A first study focused on γδ T cells as EV donors. Li et al. explored the use of γδ T cell-derived extracellular vesicles (γδ TDEs) as a drug delivery system (DDS) for miR-138 in the treatment of OSCC [109]. They found that overexpression of tumor suppressor miR-138 in γδ T cells resulted in the production of miR-138-rich TDEs. These γδTDEs could transfer miR-138 to OSCC cells, inhibiting tumor growth both in vitro and in vivo. Additionally, by targeting programmed cell death 1 (PD-1) and cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) in T-cells, the miR-138-rich γδTDEs also stimulated the proliferation and cytotoxicity of CD8+ T cells against OSCC cells. As such the miR-138-rich γδTDEs had a dual—direct and indirect—anti-tumoral effect on OSCC cells.Mesenchymal stromal cells can also function as EV-donors. Oral potentially malignant disorders (OPMDs), such as erythroplakia, oral leukoplakia, and oral submucous fibrosis, are precursor lesions that may undergo malignant transformation to OSCC [111]. A study by Wang et al. investigated the effect of engineered MSC-EVs with a high copy number of miR-185 on OPMDs development. In a dimethylbenzanthracene (DMBA)-induced OPMD model, treatment with miR-185-MSC-EVs reduced the severity of inflammation as well as the grade and number of dysplastic lesions. Furthermore, there was a significant decrease in proliferative and angiogenesis markers and miR-185-EV treatment activated the apoptotic pathway through direct targeting of Akt, an upstream regulator of caspase-9 [110].Collectively, EV-based drug delivery systems possess important advantages over conventional platforms. However, the utilization of EVs has obvious challenges as well, for example, how to load exogenous miRNAs into EVs. One method is to overexpress the desired miRNA in donor cells, followed by isolation of the miRNA-containing EVs. A second method is to load miRNAs into purified EVs directly. The main limitation of the first method is that the amount of RNA encapsulated into EVs may vary depending on the RNA species and/or sequence as well as the specific cellular mechanisms underlying RNA sorting into EVs [99]. On the other hand, direct loading of purified EVs with RNA molecules may impair their biological function as it can disrupt the EV membrane and structure [102,112]. Additionally, the use of exosomes generated in tumor cells may not be ideal for EV-based therapy as they might be carcinogenic [113]. Furthermore, the interaction of exogenous miRNAs and endogenous EVs needs to be investigated, as exemplified by the abovementioned study with γδTDEs. Nevertheless, given the intriguing and promising data, more research on EV-miRNAs as a therapeutic platform in HNSCC is warranted.Poor survival of locally advanced head and neck squamous cell carcinoma (LA-HNSCC) is partly due to challenges in early diagnosis as well as the lack of reliable biomarkers for predicting treatment outcome [6]. Currently, early diagnosis as well as staging relies on tissue biopsy and imaging studies [114,115]. As for tissue biopsy, limitations include the invasiveness of the procedure which impedes repeated sampling, as well as sampling bias due to the heterogeneity of the tumor [116]. Nowadays, liquid biopsies have become an attractive research method to identify the presence of cancer, therapy response, and cancer progression. A liquid biopsy involves the (molecular) analysis of a body fluid, most frequently blood. Advantages of liquid biopsies include the minimal invasiveness of the procedure, low cost, repeatability, and the comprehensive and real-time information on tumor cell evolution [117,118]. Liquid biopsies can be assessed for molecular biomarkers, including circulating tumor cells (CTCs), circulating tumor DNA (ctDNA), and EVs. This section will focus on EV-miRNAs as biomarkers in HNSCC diagnosis, response to therapy, prognosis, as well as its limitations and challenges. See Table 3 and Figure 2 for an overview of published studies.A lot of research has focused on the potential of EV-miRNAs as diagnostic and prognostic markers, suggesting a clinical application for HNSCC-specific EV-miRNA signatures in body fluids. As an example, higher expression of serum exosomal miR-21 and homeobox transcript antisense RNA (HOTAIR) was associated with higher clinical stage and lymph node metastasis of patients with laryngeal squamous cell carcinoma (LSCC).Moreover, the combination of increased exosomal miR-21 and HOTAIR could discriminate malignant (n = 52) from benign laryngeal disease (n = 49) with a sensitivity of 94.2% and specificity of 73.5% [119]. Similarly, another study proposed exosomal miR-21 as a diagnostic and prognostic biomarker of OSCC. It was reported that the exosomal miR-21 level was significantly higher in OSCC patients (n = 108) compared to healthy controls (n = 108) and that elevated exosomal miR-21 was associated with higher T stage and lymph node metastasis [38]. Another study indicated that the level of plasma exosomal miR-196a was higher in head and neck patients (HNC) (n = 74) compared to healthy donors (n = 30) and decreased after tumor resection, suggesting that exosomal miR-196a was released by tumor tissue. Higher exosomal miR-196a was associated with drug resistance and poor OS in patients with HNC. By determining the plasma exosomal miR-196a level, it was possible to separate patients in a chemoresistant and a chemosensitive group with a sensitivity of 85% and specificity of 70%. These data suggest that plasma exosomal miR-196a could serve as a prognostic factor and a predictor for chemoresistance in HNC patients [42]. In laryngeal squamous cell carcinoma (n = 6), RNA-seq analysis identified 34 upregulated and 41 downregulated serum exosomal miRNAs relative to healthy controls (n = 6). In the validation set (LSCC, n = 50, and 25 healthy controls), qRT-PCR revealed that miR-941-rich serum exosomes discriminated patients with LSCC patients from healthy controls with an area under the curve (AUC) of 0.797 [120]. Another study compared the miRNA content of EVs in the serum of patients with HPV positive OPSCC (n = 40), patients with gastroesophageal reflux disease (GERD; a benign inflammatory disease) (n = 20) and healthy controls (n = 20). This study used a customized miRNA-array to assess 112 miRNAs. An 11 EV-miRNA signature differentiated HPV-associated OPSCCs from healthy controls and patients with GERD with 90% sensitivity and 79% specificity [121]. More recently, Shimada et al. focused on identifying discriminating markers for differential diagnosis of a primary lung squamous cell carcinoma vs. a solitary lung metastasis from previously resected head and neck squamous cell carcinoma. The levels of miR-10a, miR-28, and miR-141 were significantly elevated in primary lung cancer compared to oligometastatic HNSCC, both in formalin-fixed paraffin-embedded (FFPE) tissue and in serum EVs. As the treatment approach and outcome differs between patients with primary early stage lung cancer vs. metastatic HNSCC, this study demonstrated that EV-miRNAs could be used as a diagnostic tool to guide clinical management [122]. In another study, a model based on EV-miR-491-5p expression was developed as a prediction tool to discriminate between patients with HNSCC and healthy controls, showing a sensitivity and specificity of 46.6% and 100%, respectively. Furthermore, it was found that the dynamic change of miRNA-491-5p pre- vs. post-treatment was associated with the 1-year disease recurrence rate (80% sensitivity and 69.23% specificity), as well as disease free survival and overall survival (HR of 2.82 and 5.66, respectively) [123]. This finding might help identify HNSCC patients who are at high risk of tumor recurrence.Saliva is the most proximal body fluid in oral cancer. As a research specimen, it has many advantages, including the fact that it is easily accessible in a noninvasive manner, it contains a low background level of normal cell material (cells, DNA, RNA, and proteins), and inhibitory substances are less abundant [132]. The potential use of saliva-derived exosomal miRNAs for the detection of HNSCC has been reported. Saliva-derived exosomal miR-512-3p and miR-412-3p were upregulated, and miR-302b-3p and miR-517b-3p selectively enriched in EVs in a study of 21 OSCC patients and 11 healthy volunteers; ROC analysis showed high diagnostic power with AUC values of 0.847 and 0.871, respectively [124]. Langevin et al. sequenced exosomal miRNAs from four HNSCC cell lines, followed by a validation study using the saliva of patients with HNSCC. The results demonstrated that the levels of miR-486-5p, miR-486-3p and miR-10b-5p are increased in saliva of HNSCC patients (n = 11) relative to healthy controls (n = 9) [125]. He et al. found that the expression level of miR-24-3p in salivary exosomes from OSCC patients (n = 4) was substantially higher than in healthy controls (n = 4) (fold change 121.54). In the validation cohort, miR-24-3p could discriminate between OSCC patients (n = 45) and healthy controls (n = 10) with a sensitivity and specificity of 64.4% and 80%, respectively [126]. In another study, miR-200a and miR-134 were significantly dysregulated in OSCC patients (n = 14) compared to smokers (n = 17) and healthy controls (n = 6) [127].Additionally, in vitro studies have identified several other EV-miRNA biomarkers in HNSCC. Erlotinib is a small-molecule inhibitor of the epidermal growth factor (EGFR) pathway, a known molecular target in HNSCC. However, there has been limited therapeutic success from EGFR inhibition in HNSCC with EGFR-targeting agents achieving response rates of about 4–15%. Almost all patients eventually develop resistance, suggesting innate and acquired resistance to EGFR inhibition [133,134,135,136]. Previous research examined the differentially expressed miRNA in EVs released from erlotinib-resistant and erlotinib-sensitive cells. In EVs released from erlotinib-resistant HNSCC cells, miR-7704, miR-21-5p, and miR-3960 were significantly upregulated. Transfection of these three miRNAs induced a pro-tumorigenic effect in cell lines. Inversely, let-7i-5p, miR-619-5p, and miR-30e-3p were downregulated in resistant cells. Transfection of these miRNAs induced an anti-tumor effect in cell lines. These results indicate that profiling of EV-miRNAs can potentially predict erlotinib response in HNSCC [128]. The same research group reported in another study that oncogenic miRNA-365 promotes OSCC cells’ progression [137] and is exported into exosomes, suggesting the potential role of EV-miRNA-365 as a biomarker for OSCC [129]. Another recent study identified dysregulated exosomal miRNAs comparing OSCC-derived cell lines HSC-2, HSC-3, Ca9-22, and HO-1-N1 and human normal oral keratinocytes (HNOKs) using an miRNA array. The four dysregulated miRNAs, miR-125b-5p, miR-17-5p, miR-200b-3p and miR-23a-3p, were reported as potential biomarkers for OSCC [130]. Small RNA sequencing was performed by Peacock et al. to identify an miRNA signature associated with EVs originating from HPV positive vs. HPV negative OPSCC cells. The analysis revealed that 14 miRNAs were enriched in EVs from HPV positive cells, while 19 miRNAs were enriched in EVs from HPV negative cells. These findings suggest that EV miRNAs could be used for oropharyngeal cancer subtype classification [131].Overall, EV-miRNA profiling provides a compelling research focus with tremendous promise. However, there are still substantial challenges to overcome before profiling of EV-miRNAs will be integrated into routine clinical use. For instance, the EV isolation and purification methods used can greatly influence the results of subsequent analyses and therefore consistency in methodology is crucial. For example, ultracentrifugation (UC) is less suitable for clinical application because the procedure may disrupt EVs when high g-forces (>100,000 g) are used [138]. Using a density gradient-based protocol typically produces purer EV fractions but is time-consuming and has a lower yield [139]. A precipitation approach produces a higher particle yield, albeit less pure with a low particle-to-protein ratio [140]. Size exclusion chromatography (SEC) has been demonstrated to retain the functional properties of EVs better than UC-isolated EVs [141]. As a disadvantage, the SEC-based procedure often requires concentration steps to concentrate the dilution of EV samples resulting from ultrafiltration.Another important challenge is the presence of significant levels of endogenous EV-miRNAs in body fluids, which hinders the identification of tumor-specific EV-miRNAs. The level of exosomal miRNAs in the circulation also fluctuates between individuals, indicating inter-individual differences. The diagnostic performance of saliva as a biomarker source appears to be tumor site-dependent, being most efficient for oral cavity cancer, as shown in ctDNA studies [114,142]. In the end, it may also be necessary to evaluate a combination of several body fluids in order to improve biomarker performance.As illustrated in this review, EV-miRNAs are involved in all aspects of tumor development in HNSCC. Despite the rapidly expanding knowledge on EV-miRNAs, a lot is still unknown regarding the exact mechanisms governing this crucial form of cell-to-cell communication. Furthermore, aside from HPV status, based on transcriptomic profiling different head and neck cancer molecular subtypes may be identified with differential involvement of EV-miRNAs in disease biology.EV-miRNAs are promising biomarkers in HNSCC, especially for (early) disease detection and prediction of treatment outcome. The limited overlap in candidate biomarker EV-miRNAs between studies is likely to reflect methodological issues in this budding field of research. Additionally, most of the studies were undertaken in small patient cohorts. Exploration and validation in large sample cohorts, in multi-center studies, using standardized protocols and analysis methods are required to prove the value of EV-miRNAs in a clinical setting. In this regard, it is necessary to include liquid biopsy analyses in clinical studies (two examples of lung cancer studies can be found on ClinicalTrials.gov: identifiers NCT04427475 and NCT03542253). To further advance the EV-miRNA field, an adequate infrastructure for collection, storage and processing of liquid biopsy specimens is needed, which can be achieved through the set-up of so-called Liquid Biopsy Centers (for example, the Cancer Center Amsterdam Liquid Biopsy Center, http://www.liquidbiopsycenter.nl/, accessed on 20 January 2022).In conclusion, EV-miRNAs hold great promise as disease biomarkers in HNSCC. Furthermore, EV-miRNAs are a likely novel therapeutic target in HNSCC. Still in the stage of promise, but now riding on the wave of the liquid biopsy revolution, EV-miRNAs will prove to be more than a bubble in the coming future.Conceptualization, W.P. and J.V.; writing—original draft preparation, W.P. and J.V.; writing—review and editing, W.P., D.M.P. and J.V. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The work was partially supported by Thailand Science Research and Innovation (Thailand Research Fund) through the Royal Golden Jubilee PhD Program and Mahidol University (PHD/0098/2559). All figures created with BioRender.com (accessed on 21 January 2022).The authors declare no conflict of interest.EV-miRNAs in HNSCC. As key players in an intricate tumor network, EV-miRNAs are involved in virtually all aspects of tumor development. NTEC, normal tongue epithelial cells; hBMSCs, human bone marrow mesenchymal stem cells; CAF, cancer-associated fibroblasts; CDDP, cisplatin; DTX, docetaxel.Clinical applications of EV-miRNAs in HNSCC. Upper panels display EV-miRNAs as a drug delivery system for miRNA-based therapy. The lower panel provides an overview of EV- miRNA-based candidate biomarkers for diagnosis, prognosis, and treatment response in HNSCC. CTLA-4, cytotoxic T-lymphocyte-associated antigen 4; MSC, mesenchymal stromal cell; OPMD, oral potentially malignant disorders; OSCC, oral squamous cell carcinoma; PD-1, programmed cell death 1.EV-miRNAs in HNSCC.Abbreviations: AXL, AXL receptor tyrosine kinase; BCL2, B-cell lymphoma 2; CAF, cancer-associated fibroblast; CAV1, Caveolin-1; CDDP, cisplatin; CDKN1B, cyclin-dependent kinase inhibitor 1B; CHD9, chromodomain helicase DNA binding protein 9; COL10A1, collagen type X alpha 1 chain gene; DENND2D, DENN/MADD Domain Containing 2D; DTX, docetaxel; hBMSCs, human bone marrow mesenchymal stem cells; HNSCC, head and neck squamous cell carcinoma; HUVEC, human umbilical vein endothelial cell; IL12A, interleukin 12A; ING5, inhibitor of growth 5; NGF, normal gingival fibroblast; OSCC, oral squamous cell carcinoma; PDCD4, programmed cell death protein 4; PPARδ, peroxisome proliferator-activated receptor δ; PPP2R1B, protein phosphatase 2 scaffold subunit Abeta; PTEN, phosphatase and tensin homolog; TSCC, tongue squamous cell carcinoma; TUBB3, tubulin beta 3 class III; WRN, Werner syndrome RecQ like helicase.EV-miRNA as a drug delivery system in HNSCC.Abbreviations: CTLA-4, cytotoxic T-lymphocyte-associated antigen 4; MSC, mesenchymal stromal cell; OPMD, oral potentially malignant disorders; OSCC, oral squamous cell carcinoma; PD-1, programmed cell death 1.EV-miRNAs as disease biomarkers for HNSCC.Abbreviations: CDDP, cisplatin; GORD, gastroesophageal reflux disease; HNSCC, head and neck squamous cell carcinoma; LSCC, laryngeal squamous cell carcinoma; LSQCC, lung squamous cell carcinoma; MSQCC, solitary metastatic lung tumor; OPSCC, oropharyngeal squamous cell carcinoma; OSCC, oral squamous cell carcinoma.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01161.txt

@@ -0,0 +1,14 @@
+Members of the PelvEx Collaborative could be found in the Appendix A.Correspondence to Michael Eamon Kelly, Department of Colorectal Surgery, St James Hospital, D8 Dublin, Ireland; kellym11@tcd.ie.Pelvic exenteration is a complex procedure performed for the management of advanced pelvic cancers. It often involves the resection of several pelvic organs and can be associated with a high morbidity and impact on the patient’s quality of life. The development of better diagnostics, improved chemotherapy and radiotherapy regimens, combined with advanced surgical strategies have improved surgical and survival outcomes. This article highlights current and future management options.Pelvic exenteration is a complex operation performed for locally advanced and recurrent pelvic cancers. The goal of surgery is to achieve clear margins, therefore identifying adjacent or involved organs, bone, muscle, nerves and/or vascular structures that may need resection. While these extensive resections are potentially curative, they can be associated with substantial morbidity. Recently, there has been a move to centralize care to specialized units, as this facilitates better multidisciplinary care input. Advancements in pelvic oncology and surgical innovation have redefined the boundaries of pelvic exenterative surgery. Combined with improved neoadjuvant therapies, advances in diagnostics, and better reconstructive techniques have provided quicker recovery and better quality of life outcomes, with improved survival This article provides highlights of the current management of advanced pelvic cancers in terms of surgical strategy and potential future developments.Pelvic exenteration is a complex operation performed for locally advanced and recurrent pelvic cancers [1]. The procedure involves an the en bloc resection of at least two pelvic organs with subsequent reconstruction and/or diversion of bowel/urinary/sexual functions. The goal of surgery is to achieve clear margins; therefore, adjacent or involved organs, bone, muscle, nerves and vascular structures may be resected [2,3]. While these extensive resections are potentially curative, they are associated with significant morbidity [4]. A multidisciplinary team approach is essential to optimise patients pre-operatively and during their recovery process [5]. More recently, there has been a move to centralize care to specialized units, as this facilitates a better integration of prehabilitation protocols, subspecialty involvement and a greater emphasis on research and quality of life assessments [5,6,7,8].The COVID-19 pandemic has disrupted patient access to healthcare worldwide, despite efforts to maintain essential cancer care during this time [9]. Early studies have shown a concerning delay in cancer diagnosis and treatment during the initial ‘waves’ of the pandemic [10,11]. This is predicted to cause an increase in the proportion of patients with more advanced stages of cancer [9]. This may challenge our already strained healthcare services, with protected surgical beds and the availability of intensive care facilities essential for maintaining complex surgical care [12].Advancements in pelvic oncology and surgical innovation have redefined the boundaries of pelvic operations [13]. Aggressive surgical techniques including extended pelvic exenteration (bony/vascular resection) and cytoreductive surgery with or without hyperthermic intraperitoneal chemotherapy (CRS-HIPEC) reported improved five-year overall survival rates [14,15,16,17,18]. Improved neoadjuvant therapies and advances in imaging techniques, navigational technology and artificial intelligence facilitate the increased downstaging of advanced neoplasms, more judicious patient selection and/or greater surgical precision [19,20]. Reconstructive techniques provided quicker recovery and better quality of life outcomes [21]. This article aims to provide insight into the contemporary management of advanced pelvic cancers in terms of surgical strategy and future developments.Precise tumour staging is essential to determine prognosis and enables a more focused assessment of the available management options [13]. Local staging for rectal cancer is most accurately established with magnetic resonance imaging (MRI), with a sensitivity and specificity of up to 100% and 98%, respectively, for T3 and T4 disease. Staging for distant metastases is routinely performed with computerised tomography (CT) of the thorax, abdomen and pelvis [14]. Recently, the concept of whole-body MRI (WB-MRI) has been postulated, as a radiation-free alternative for surveying cancer patients. Furthermore, WB-MRI is shown to be highly accurate in the detection of bone metastases. Despite this, WB-MRI has not been widely utilized [15,22]. Ongoing studies are comparing it to positron emission tomography (PET-CT), especially in the setting of mucinous adenocarcinoma, peritoneal malignancy and other PET-insensitive neoplasms [15,23].The use of MRI of the chest/thorax is not widely available and is subject to significant operator variation [16]. However, the role of MRI of the abdomen/pelvis is well-established [24]. MRI liver is used to characterise equivocal lesions on CT [25]. Therefore, the role of WB-MRI may evolve over time as a viable alternative to CT.Regardless of modality, good communication and feedback between surgical and radiological departments is critical in the assessment of patient operability and for predicting resectability. Expertise is required to devise a radiological roadmap that incorporates surgical planes, highlighting potential issues and ensuring negative margins. A roadmap for R0 excision should be tailored to the maximum possible disease extent previously identified on sequential MRI imaging, regardless of down-staging post-neoadjuvant treatment. The rationale behind this is based on the knowledge that radiologically occult microscopic foci of viable tumour cells can occasionally persist beyond down-staged tumour margins (peri-tumoral scar tissue). For this reason, areas of fibrosis in contact with tumours on post-treatment imaging should be considered to have malignant potential and included in the extended resection [26,27,28,29,30]. The radiological roadmap should be tailored individually to each patient, accounting for their anatomy, tumour extent and co-morbidities. ‘BONVUE’ is a helpful acronym used to ensure the radiology team includes a description of bones, organs, nerves, vessels, ureters and extra (tumour sites) in the development of a roadmap [31].Total neoadjuvant therapy (TNT) is increasingly utilized for the downstaging of locally advanced rectal cancer (LARC) [32]. Recent studies highlight that TNT may improve outcomes by increasing patient compliance to therapy, reducing tumour stage, and exposing patients to earlier chemotherapy [19]. TNT strategies vary between centres, with some advocating chemo-induction prior to long-course chemoradiotherapy (induction), while others favour consolidation chemotherapy (long-course chemoradiotherapy followed by chemotherapy) [33]. Two recent meta-analyses demonstrated an improved pathological complete response (pCR) in patients who underwent TNT compared to conventional treatment [19,32]. Recently, the PelvEx Collaborative has become involved in two randomised controlled trials examining the role of TNT in the setting of recurrent rectal cancer. PelvEx II trial (NCT 04389088) and GRECCAR 15 have started to recruit patients, and their outcomes will influence the future management of patients with recurrent rectal cancer. The PelvEx II trial is a multi-centre, open-label, randomised, controlled, parallel arms clinical trial of induction chemotherapy followed by chemoradiotherapy versus chemoradiotherapy alone as neoadjuvant treatments for locally recurrent rectal cancer (LRRC). Similarly, the GRECCAR 15 trial is a phase III randomised trial aiming to evaluate chemotherapy followed by pelvic re-irradiation versus chemotherapy alone as neoadjuvant treatment in LRRC [34].The treatment response to neoadjuvant chemoradiotherapy is highly variable, with up to 20% of rectal tumours exhibiting a complete resistance [35]. The activation of mutations in genes in the phosphatidylinositol 3-kinase (PI3K) and MAP kinase (MAPK) signalling pathways is shown to modulate treatment response and clinical outcomes in locally advanced rectal cancers [36]. Gene mutations associated with an increased response to TNT include ARID1A, PMS2 and JAK1, whereas those associated with resisting treatment include KDM6A, ABL1 and DNMT3A [35]. In addition, work on the relationship between treatment resistance and the microbiome is emerging. One study found greater deposits of fusobacteria in an RNA analysis of pre-treatment tumours in intermediate and poor responders to neoadjuvant therapies [36]. This opens several promising avenues that can be investigated, especially as genomic sequencing increasingly influences the preferred neoadjuvant regimen or agent.Programmed cell death-1 (PD-1) is a cell receptor found on the surface of activated T-cells, pro-B cells and macrophages that contains at least two ligands; programmed death-ligand 1 (PD-L1) and programmed death-ligand 2 (PD-L2). The binding of these two ligands to the PD-1 receptor results in T-cell deactivation and subsequent tumour cell evasion, preventing a host attack on its own immune system [37]. PD pathway blockade, in order to prevent immune evasion, is a novel method that can also be considered [37,38]. A recent meta-analysis revealed prolonged survival rates in patients with dMMR/MSI-H mCRC receiving anti-PD-1 inhibitor monotherapy [37]. Despite these promising results, many questions remain. dMMR/MSI-H tumours only account for 5% of mCRC and further research is required to extend the benefit of immunotherapy into a broader, microsatellite stable population [39].Perioperative strategies to optimise the outcomes of patients undergoing pelvic exenteration or extended resection for pelvic cancers is critical to maximising treatment success [40]. Pre-existing co-morbidities are associated with poorer outcomes [25]. Prehabilitation can be defined as the process of ��optimising physical functionality preoperatively to enable the individual to maintain a normal level of function during and after surgery’. It encompasses a combination of exercise, nutrition and psychosocial interventions [41,42]. A meta-analysis of fifteen studies revealed a significantly lower hospital length of stay in patients undergoing cancer surgery, demonstrating an accelerated post-operative recovery in patients exposed to prehabilitation in the pre-operative period [42]. To optimise patient outcomes, multidisciplinary collaboration is essential, incorporating opinions from members of the anaesthetic and surgical teams, nursing staff and other allied health professionals [40].Surgical patients with a poor functional capacity, determined by oxygen consumption at anaerobic threshold (AT) during cardiopulmonary exercise testing (CPET), experience poorer post-operative outcomes. The identification of high-risk surgical patients allows for the appropriate planning of their perioperative care, subsequently reducing the risk of mortality or severe complications in the post-operative period [43]. Pre-operative cardiopulmonary exercise testing (PCPET) allows us to assess exercise capacity whilst identifying causes of exercise limitations. Information acquired from PCPET can be invaluable for estimating the risk of perioperative events [44]. Anaerobic threshold (AT) indicates the status of the patient’s aerobic fitness and is predictive of perioperative outcomes [43]. The PelvEx Collaborative previously outlined five consensus recommendations to optimise preoperative assessment and preparation in patients undergoing pelvic exenteration [40]:Where possible, the anaesthetist undertaking the case should personally pre-assess the high-risk patient undergoing exenteration;When available, CPET should be utilised to assess functional capacity pre-operatively;Pelvic exenteration can be undertaken in patients who have demonstrated an adequate CPET result and have been deemed low risk for severe perioperative morbidity;Patients with more than two cardiac risk factors and poor functional capacity should undergo imaging stress-testing prior to surgery;Formal cardiology assessment is not routinely required in patients undergoing pelvic exenteration.Cancer-related malnutrition occurs secondary to anorexia, nausea, vomiting, metabolic disorders and psychological factors in patients undergoing major oncological surgery. Two cohort studies identified that 32.5% and 24% of their populations, respectively, were malnourished before exenteration when assessed by the subjective global assessment (SGA) tool [25]. The optimisation of nutritional and metabolic state prior to surgery contributes to improved perioperative outcomes and is being increasingly employed as part of pre-operative MDT disease management [45]. Malnutrition during neoadjuvant therapy was also associated with adverse perioperative outcomes, including reduced tumour response, poor treatment tolerance and increased morbidity [46,47]. Early identification and treatment of malnutrition was shown to improve outcomes, lower infection rates, shorten hospital stay and improve wound healing [48].Negative resection margins (R0) are the single most important prognostic factor in predicting long-term survival in patients undergoing pelvic exenteration [1,5,7,8,13,49,50,51,52,53,54,55]. The goal of exenterative surgery is to resect all involved organs/structures whilst balancing this radicality with an acceptable risk profile and postoperative quality of life. In recent decades, more extensive procedures are being performed, with better patient education and counselling regarding the risks [14,17]. Various surgical techniques have been developed to facilitate en bloc resection of ‘higher and wider’ pelvic tumours beyond traditional mesorectal planes. These include high sacrectomy, pubic bone resection and lateral compartment excision, often involving major neurovascular structures [17]. Low sacrectomy (below S3) is performed routinely by exenterative surgeons, demonstrating relatively low complication rates regardless of patient positioning [13]. Similarly, high sacrectomy is shown to be safe and efficacious without compromising R0 rates and is no longer considered a contraindication to surgery [56]. Several studies showed that an R0 resection can be achieved in 55–80% of patients with recurrent rectal cancer undergoing exenterative surgery, translating to a 5-year overall survival of 28–50% [57,58,59]. Several specialised units have adopted novel techniques for en bloc sacral resection that minimise morbidity by avoiding complete sacrectomy. Proposed methods include anterior sacrectomy (resection of the anterior cortex to preserve nerve roots), segmental sacrectomy or high subcortical sacrectomy (HiSS) [60,61]. In the past, the involvement of the pelvic sidewall was considered an absolute contraindication for surgery due to bony limitations and the presence of major neurovascular structures [54]. However, better reconstructive methods have allowed for these more radical resections [62]. Increasingly, the selective en bloc resection of the pelvic side-wall structures, including the internal iliac vessels, piriformis and obturator internus muscles, ischium, and sacrotuberous/sacropspinous ligaments is being performed [63].Historically, the presence of hydronephrosis, gross lower limb oedema, and invasion of the sciatic notch or involvement of the aortoiliac axis suggested inoperable disease. Various studies have demonstrated the safety and efficacy of major extra-anatomic resections involving these structures in selective patients [64,65,66]. En bloc sciatic nerve and/or lumbosacral trunk resections for tumours extending laterally into the piriformis muscle have demonstrated similar R0 rates to central pelvic tumours [13]. Functional outcomes are always a concern when undertaking these radical resections; however, almost all patients undergoing complete sciatic nerve resection regain mobility post-operatively following intensive physiotherapy and orthotics input [64]. En bloc major vascular resections of the aortoiliac axis are also shown to be feasible in select patients in specialised centres, with an R0 rate of 81.8% reported in one study [65].While the extent of resection is theoretically limitless, it is imperative that morbidity is minimized and, therefore, operations are tailored to each patient. As advancements in reconstructive methods and rehabilitative systems are made, the indications and contraindications for surgery are constantly changing [14]. Currently, absolute contraindications to resection include: poor performance status, bilateral sciatic nerve involvement and/or circumferential bone involvement. Relative contraindications include: encasement of external iliac vasculature, high sacral involvement (above S2), extension through the sciatic notch, and/or unresectable extra-pelvic metastases [30].IORT delivers a single high-fraction dose (10–20 Gy) of radiation directly to anatomical targets deemed as having the potential for high recurrence risk [67]. IORT is usually administrated to patients with either no or a limited volume of metastatic disease [68]. The addition of IORT to conventional multi-modal treatment strategies has been shown to achieve excellent local control outcomes in ‘select’ patient cohorts [67,69]. The utilization of IORT in the setting of gross residual (R2) disease is of limited value.IORT can be delivered via two methods: high-dose-rate brachytherapy or intraoperative electron beam radiotherapy (IOERT). The evidence for IORT stems from positive prospective long-term data from several international units who administer it when there is concern for threatened margins [70]. A recent systematic review suggested that IORT may improve oncological outcomes in advanced and recurrent colorectal cancers, offering better local control, disease-free survival and overall survival with no associated increase in severe complications [67].Current treatment options for patients with peritoneal metastases include supportive care, palliative systemic chemotherapy, pressurised intraperitoneal aerosol chemotherapy (PIPAC), and cytoreductive surgery (CRS) combined with heated intraperitoneal chemotherapy (CRS-HIPEC). When treated with modern systemic chemotherapy, these patients have a median survival of thirteen months [71]. Patients with low-volume peritoneal disease may be considered for CRS-HIPEC with curative intent, whereas those with more extensive disease may benefit more from PIPAC [72].CRS-HIPEC is a well-established treatment modality for patients with synchronous or metachronous colorectal peritoneal metastases [73]. While it is associated with a long-term survival benefit, high rates of morbidity were observed to range from 12 to 65% [74]. Traditionally, CRS-HIPEC is not recommended in those needing a pelvic exenteration [75]. Recent studies suggest the feasibility and safety of these two procedures being performed simultaneously, with an acceptable level of morbidity [74,75]. The PRODIGE-7 trial compared CRS and CRS-HIPEC in patients with peritoneal metastases. All patients received at least six months of oxaliplatin-based systemic chemotherapy, with the CRS-HIPEC arm receiving additional HIPEC with oxaliplatin for 30 min. Median survival was 41.7 and 41.2 months with and without HIPEC, respectively [76]. This study demonstrated that oxaliplatin-based HIPEC did not improve survival in this cohort of patients. Tuech et al. recently demonstrated a complete cytoreduction in all patients undergoing CRS-HIPEC and TPE, with R0 margins achieved in 81.2%. Despite these promising results, severe complications occurred in 56.2% of patients and post-operative mortality was 12.5% [74].Further research is required via a multi-centre approach to determine the optimal candidates for this approach [77]. The PRODIGE 7 trial demonstrated no clear benefit in the use of oxaliplatin-based HIPEC in addition to standard chemotherapy. However, more data are required to evaluate if other HIPEC chemotherapeutic regimens are better.For patients presenting with rectal cancer, 15–20% will have synchronous liver metastases [52]. The optimal management of these patients is subject to debate and often dependent on local resources and expertise [51]. Historically, surgical resection in patients with LARC or LRRC was confined to patients without metastatic disease. However, simultaneous hepatic resection was reported to be technically feasible with an acceptable morbidity and mortality rate when performed on select patients [52,78]. The median cancer-specific survival after liver resection for colorectal cancer with liver metastases was reported as 42.5 months, with disease tending to recur in patients with poor differentiation of the primary tumour, positive lymph nodes and higher amounts of liver metastases [79].The PelvEx Collaborative observed an R0 resection in 73.5% of pelvic exenterations and 66.4% of liver resections among 128 patients with synchronous liver metastases. The 5-year overall survival for patients in whom an R0 resection was achieved was 54.6% in comparison to 20% for those with an R1/R2 resection. This was the first multi-centre study that demonstrated the safety and feasibility of simultaneous liver resection, with acceptable morbidity and mortality rate [52].Debate around the optimal treatment of colorectal lung metastases also remains. [80]. The majority of these metastases are suitable for surgical resection, with reasonable 5-year overall survival rates [81]. A recent meta-analysis demonstrated comparable survival rates between both the surgical and non-surgical management of colorectal pulmonary metastases, contrary to earlier evidence suggesting a benefit of resection [80]. The recently published LaIT-SABR study aimed to identify predictive factors of sterotactic ablative radiotherapy (SABR) response in patients with colorectal lung metastases and investigate the rates of progression to polymetastatic disease [82]. Their results support the use of SBRT in this cohort of patients as it was associated with a delay in progression to polymetastatic disease. Further prospective studies are necessary to obtain a better understanding of the long-term effect of lung metastasectomy in metastatic colorectal cancer [83].Minimally invasive surgical modalities have evolved considerably in recent years, particularly regarding pelvic procedures [84]. A recent meta-analysis published by the PelvEx Collaborative investigated the current evidence regarding the use of MIS techniques such as laparoscopy and robotic surgery in pelvic exenteration. It was concluded that MIS exenteration was associated with reduced intra-operative blood loss and hospital length of stay while having no adverse effect on resectability [1]. Since robotic-assisted pelvic exenteration was first described in 2013, there are increasing numbers of case reports and series demonstrating its safety and feasibility [85]. The current evidence in the literature suggests an acceptable operative time, blood loss and a range of R0 rates [84,86,87,88,89,90,91]. Laparoscopic pelvic exenteration has been associated with reduced blood loss, faster recovery and an acceptable length of stay; on the contrary, in well-selected patients, the learning curve is steep [92]. Robotic-assisted surgery facilitates a more ergonomic and visually enhanced platform [93].In the PelvEx Collaborative meta-analysis comparing MIS techniques to the open approach, 78.1% underwent open exenteration while 21.8% had an MIS exenteration analysis among 170 patients. MIS exenteration was associated with a longer operating time but substantially less blood loss. MIS exenteration was also associated with a significantly reduced overall morbidity rate (56.7% versus 88.5%) and a short post-operative length of stay (6 days less). This study demonstrated the safety and feasibility of MIS exenteration in patients with favourable anatomy and tumour characteristics [1]. Moving forward, novel robotic technology such as fluorescence-guided surgery, 3-dimensional modelling and stereotactic navigation will significantly improve surgical dissection and resection margins [93]. Fluorescence-guided surgery was established in several different specialities. Indocyanine green (ICG) can be used to map lymph nodes in various cancers, detect tumour margins and evaluating bowel perfusion at anastomotic sites [94]. Three-dimensional modelling with a virtual reality viewing system is shown to augment the surgical planning process and result in improved patient outcomes. This involves the conversion of CT and MRI images to 3-dimensional virtual reality models helping the surgeon plan the operation [95]. Stereotactic navigation paired with robotics is a novel concept but remains technically challenging. The addition of navigation to robotics will undoubtedly improve surgical precision [96].Radical resections incur a greater need for reconstruction. The ability to perform complex soft tissue, vascular and bone reconstruction/stabilization has improved the functional outcomes of patients undergoing pelvic exenteration.Many patients will require flap reconstruction after exenterative surgery due to extensive tissue loss. However, some co-existing factors will determine the feasibility of each specific reconstruction. Previous chemoradiotherapy, increased pelvic dead space, poor tissue vascular supply, accumulation of fluid and bacterial contamination all play a role in the development of flap complications, which occur in 25–60% of reconstructions [97]. The following soft tissue reconstruction methods can be considered, based on certain regions [98,99]:
+              
+                Abdominal: Vertical or oblique rectus abdominus myocutaneous (VRAM/ORAM).
+              
+              
+                Gluteal: Myocutaneous or fascio-cutaneous VY-plasty, inferior gluteal artery perforator (IGAP) flap.
+              
+              
+                Upper thigh: Anterolateral thigh +/− vastus lateralis flap, bilateral pedicled gracilis flap.
+              
+              
+                Gluteal fold/perineal: Internal pudendal artery perforator or perineal turnover perforator flap.
+              
+            Abdominal: Vertical or oblique rectus abdominus myocutaneous (VRAM/ORAM).Gluteal: Myocutaneous or fascio-cutaneous VY-plasty, inferior gluteal artery perforator (IGAP) flap.Upper thigh: Anterolateral thigh +/− vastus lateralis flap, bilateral pedicled gracilis flap.Gluteal fold/perineal: Internal pudendal artery perforator or perineal turnover perforator flap.Vaginal defects resulting from radical oncologic resection are challenging to reconstruct. These defects may range from simple mucosal defects to full circumferential loss due to posterior vaginal wall resection. Anatomy and function can be restored using a rectus abdominis myofascial flap, deep inferior epigastric perforator (DIEP) flap, bilateral gracilis flats or gluteus maximus special flaps [25].Empty pelvis syndrome is a major contributor to morbidity following pelvic exenteration as dead space allows for the accumulation of fluid and small bowel migration (obstruction) into the pelvis. To alleviate the risk of these complications, the dead space must be “filled” using either synthetic mesh or tissue. Reconstructive methods include myocutaneous flap reconstruction, omental flaps and mesh reconstruction. There is currently insufficient evidence in the literature to support the use of one reconstructive method over another [100].En bloc sacrectomy is performed in cases where tumours infiltrate the presacral fascia and may require further reconstruction [17]. Sacrectomy results in a large cavity which can result in infection as well as neurological or sexual deficits. Reconstruction aims to restore the pelvic ring and spinopelvic junction [18]. While several fixation methods exist, such as spinopelvic fixation (SPF), posterior pelvic ring fixation (PPRF) and anterior spinal column fixation (ASCF), there is a lack of evidence to suggest the superiority of one method over another [97].The involvement of aortoiliac vessels, the sciatic nerve or its associated roots substantially increases the difficulty of achieving an R0 resection in advanced pelvic malignancies [17]. To achieve a clear lateral margin, iliac vessels may be resected en bloc and subsequently reconstructed with an autologous or synthetic graft [17]. A pre-emptive femoral–femoral arterial and venous crossover graft reconstruction method has also been studied and demonstrated a decreased risk of graft infection secondary to avoidance of contamination with gastrointestinal or genitourinary organisms [101,102].Palliative procedures, such as a defunctioning stoma or urinary diversion, can be performed in patients who are suffering from disabling symptoms where an R0 resection is unlikely [49]. The decision to undergo major palliative surgery in the setting of an advanced/recurrent rectal must be considered on a case-by-case basis [103]. Locally advanced disease can have a profound impact on a patient’s QoL. Relentless growth can cause intractable symptoms, including pain, bleeding, fistulisation to bladder/abdominal wall/bone and/or intestinal obstruction [104]. Increasingly, palliative exenteration is being considered. This remains a controversial topic but should be discussed with patients so that they understand all risks and benefits [49]. Clear counselling, with the involvement of the multidisciplinary team, is vital for establishing treatment goals and expectations [105]. Quyn et al. observed a 62% response rate to their QoL questionnaire using the Functional Assessment of Cancer Therapy—Colorectal (FACT-C). The average FACT-C score returned to pre-operative quality of life two months post-operatively and quality of life continued to improve slowly over the following twelve months [3].A recent meta-analysis performed by the PelvEx Collaborative demonstrated symptom relief in ‘select patients’ undergoing palliative exenteration. Symptom relief was reported in a median of 79% of patients, although the magnitude of the effect was poorly measured. Though available data are limited (23 studies comprising 509 patients), the results suggest that there may be some improvement in symptom control in selective patients. However, palliative exenteration is an extremely morbid procedure with insufficient evidence of sustained quality of life [49]. As the median overall survival was only 14 months in this cohort of patients, it is essential to consider safer procedures such as stoma formation, bypass, nephrostomy, radiotherapy and multimodal analgesia before resection.Enhanced recovery after surgery (ERAS) protocols are well-established and have demonstrated improvements in morbidity rates, length of stay and quality of life [106]. There is emerging evidence to suggest the feasibility of ERAS in complex cytoreductive surgery with an improvement in early clinical outcomes [107,108,109]. The PelvEx Collaborative has offered guidance on the perioperative management of patients undergoing exenterative surgery and acknowledges the need for individualised tailored post-operative treatment plans [40].Harji et al. enrolled 145 patients into a dedicated pelvic exenteration ERAS programme to assess its feasibility and short-term impact on this cohort of patients. They demonstrated an overall compliance rate of 70%. Patients with higher compliance to the program tended to have a shorter hospital length of stay, reduced rate and severity of post-operative morbidity, as well as fewer readmissions. ERAS appears feasible and efficacious in patients undergoing pelvic exenteration, displaying a high compliance and improved clinical outcomes [110]. See Figure 1 for the highlights of contemporary management strategies.The management of advanced colorectal malignancies continues to evolve rapidly, with the introduction of new technologies, pushing the boundaries of surgical resection and involvement in clinical trials and collaborative research. Increasingly, the focus of treatment is balanced between cure and quality of life [111]. The IMPACT initiative highlighted the importance of patient involvement in the decision-making process, incorporating functional and quality outcomes as ‘key’ measures of oncological success [112]. This focus on patient-centred care, combined with burgeoning diagnostic and technological advancements, will continue to shape the approach to colorectal cancer as we move into the era of “personalized medicine”.Radiomics and radiogenomics are being investigated as a novel way to analyse images and to increase the precision of diagnostics [113]. The incorporation of artificial intelligence with biomarkers will allow clinicians to predict treatment response and may help to personalize care [113,114]. Radiomics can highlight tumour properties throughout serial imaging, and in sufficiently large datasets can uncover previously unknown markers or patterns of disease progression and/or response to chemoradiation [114].The radiomics analysis of contrast-enhanced CT images has already demonstrated improved accuracy for nodal assessment in advanced rectal cancer [115]. Similarly, a pre-treatment MRI-based machine learning model was developed which can accurately predict treatment response in patients with LARC [116]. These novel technologies may allow for a tailored approach to the treatment of advanced malignancies and help select patients with previously unrecognized adverse features that would benefit from more conservative treatment modalities.Radiogenomics is the extension of radiomics by its combination with molecular analysis in the form of genomic and transcriptomic data. Genetic analysis remains expensive, invasive, and time-consuming. Radiogenomics may play a vital role in providing imaging surrogates that correlate with genetic expression, thereby providing an alternative to genetic testing [117]. Of course, larger prospective studies with standardization are needed to validate this area of research.The adenoma-carcinoma sequence was first described in 1990 by Vogelstein and Fearon and provided the foundation for our understanding of colorectal cancer as a disease consisting of complex genomic changes [118]. Comparative genomic hybridization (CGH) arrays, single-nucleotide polymorphism (SNP) arrays and novel next-generation sequencing (NGS) approaches have provided us with insights into the complex colorectal cancer genome [119]. Subsequent research has refined and expanded our knowledge of this area, allowing us to incorporate genomics into our treatment choices. The impact of the 100,000 genome project and the integration of genomic and translational medicine into cancer care pathways has provided a unique opportunity for tailoring and personalizing oncological treatment [120]. Research suggests that, in the future, we will be able to predict responses to chemotherapeutics, which will undoubtedly guide decision making, particularly in borderline cases [121]. While the effect of RAS and BRAF mutations is well-established in current clinical practice, new genomic markers are showing promising results in clinical trials [118].BRAF V600E is the most common potentially targetable mutation in metastatic colorectal cancer; however, RAF inhibitors have limited efficacy as single agents in treating patients with this alteration [118]. The BEACON trial is currently comparing doublet or triplet targeted therapy with standard therapy in patients with BRAF V600E metastatic colorectal cancer [122]. Preliminary results are promising, with a 48% overall response rate in patients receiving triplet therapy consisting of encorafenib (RAF inhibitor), binimetinib (MEK inhibitor) and cetuximab [122].HER2 amplification occurs in 2–6% of metastatic colorectal cancers and is associated with a poor response to EGFR antibody treatment [118]. While anti-HER2 drugs, such as trastuzumab as monotherapy, have not demonstrated tumour regression in clinical trials, a combination therapy with an EGFR inhibitor achieved tumour shrinkage [123]. The phase 2 HERACLES trial reported a 30% response rate in patients with HER2-positive metastatic colorectal cancer receiving this combination therapy [124]. Cohort B of the HERACLES trial is ongoing and aims to compare the efficacy of the antibody-drug conjugate TDM-1 monotherapy with combination therapy of TDM-1 and pertuzumab (anti-HER2) in the second line setting [118].Our understanding of tumour biology and molecular subtypes is constantly expanding, thanks to advancements in microarray and NGS technologies which allow for the identification of new cancer genes and pathways. While translational genomic studies have already provided clinically relevant biomarkers for predicting prognosis and therapy response, future research will identify new drug targets and reveal novel therapeutic opportunities. Innovations in current applications, coupled with novel emerging technologies, will lead to further advancements in translational cancer genomics which will hopefully contribute to improved patient outcomes in the future [119].Over the last two decades, experienced exenterative surgeons have redefined what constitutes resectable disease. The development of regional/national specialized units has allowed funding and structural/service supports to enable these centres to establish specialist pelvic oncology units. As a result, extended bony and/or neurovascular resections are pursued more frequently, with an acceptable morbidity reported [18,65,102].‘Higher and wider’ resections at the periphery of the pelvis are now commonplace in pelvic exenteration [13]. Certain centres are performing en bloc sciatic nerve and/or lumbosacral trunk excision for tumours that extend laterally into the piriformis. High sacrectomy (above the junction of S2/S3) can be performed without compromising margins, and functional outcomes are acceptable [125]. Alternative techniques such as anterior cortical sacrectomy and abdominolithotomy sacrectomy have become more standardized.Improved reconstructive techniques have facilitated these more radical resections [21,65,98,99]. However, the repair of large bony defects remains a challenge [126]. The current methods of reconstruction of these defects include autologous iliac grafting, autologous vascularized fibula transplantation, Masquelet’s induced membrane or massive allografts. Autologous grafts account for approximately 50% of cases [126,127]. 3D bioprinting is a state-of-the-art technology used to build constructs from a single-cell type using a layer-by-layer deposition of a specific bioink [127]. Bioprinting uses cell-laden hydrogens to print structures following a period of maturation which can be developed into a variety of complex tissues [126]. Its use in bone reconstruction is still evolving and the clinical application of this technology remains in its infancy. Despite this, bio-printed bone has been successfully implanted in pre-clinical models and other 3D-printed materials have been successfully transplanted into human subjects [128]. This ground-breaking technology will allow us to develop tailored bone grafts that incorporate real cells, growth factors and vasculature, which may revolutionize the way we reconstruct bony defects in the future [126].Anatomical reconstruction of the sacrum using 3-dimensional printing technology has been sparsely reported in the literature [129,130]. Kim et al. successfully reconstructed the sacrum with a 3D-printed implant in a patient who had undergone hemispherectomy for sacral osteosarcoma. One-year follow up revealed excellent bony union without complication, demonstrating the feasibility of this novel method [129]. Similarly, Chatain et al. described a case of custom 3D-printed sacral implant for revision of failing sacrectomy in a patient who previously underwent en bloc sacrectomy and standard spinopelvic reconstruction for sacral chordoma [130]. In this case, the prosthesis was fashioned from titanium alloy using a 3D-printing technique, tailored to the patient using a CT 3-dimensional reconstruction model. The surgical implantation of the device proved challenging but long-term outcomes were satisfactory [130].Current options for tissue reconstruction rely heavily on autologous donor tissue to repair defects. 3D bioprinting offers the potential to avoid autologous grafts and the complications associated with them [131]. Its use has now been reported in a variety of surgical disciplines, including plastics, breast, orthopaedic, craniomaxillofacial and head/neck oncology [132]. While significant advancements are being made in the production of simple, single tissue types, composite tissue engineering consisting of multilaminar constructs adds an additional layer of complexity [133]. Ultimately, 3D bioprinting has the potential to produce patient-specific body components, including organs and limbs, which will undoubtedly revolutionize surgery [131].Magnetic resonance-guided histopathology has been shown to increase the accuracy of staging in LARC [134]. The technique is performed by carefully matching multilevel histologic sections, using previous MR images as guidance, to examine for evidence of residual tumour, while paying particular attention to areas with MRI signals consistent with fibrosis or mucin. Another advancement in the field of histopathology is the use of whole-slide imaging, which produces digital histologic images from glass slides [135]. This technology will allow for the precise evaluation of tumour dimensions, stage and margins, and has the potential to improve both diagnostic accuracy and workflow efficiency in the future.Biobanking is the process of collecting and storing various human specimens for the purpose of clinical research and provides a platform for the development of translational and personalised “precision” medicine [136]. Translational research with specimens from tissue biobanks enables the discovery of molecular biomarkers that have the potential to guide therapy and individualize treatment [137,138]. Many research programmes have benefited from biobank specimens, including the development of trastuzumab [139]. More recently, biobanks played a crucial role in the creation of The Cancer Genome Atlas (TCGA), a comprehensive catalogue of cancer genomic profiles [140]. This atlas has allowed for the discovery of molecular aberrations at DNA, RNA, protein and epigenetic levels, providing a detailed analysis of commonalities and differences across tumour lineages.Biobanks provide researchers with human specimens and associated clinical data, which allow for large cohorts of over 30 specimens to be analysed using large-scale genome sequencing. This facilitates the discovery of novel molecular alterations as well as the classification of tumour subtypes according to distinct genomic alterations, providing a personalised, precision medicine approach in cancer care [141].Aggressive multi-visceral resections are performed more often to manage patients with advanced pelvic cancers [111]. Currently, surgery remains the only long-term curative option in the majority of cases [142]. Despite the radicality of pelvic exenteration, previous studies have demonstrated an acceptable survival rate with reasonable quality of life outcomes [143]. It appears that quality of life scores rapidly deteriorate in the immediate post-operative period; however, they begin to rise slowly again from three months post-operation [144]. The ultimate goal of therapy is to balance patient quality of life with survival and complication rates, and this should be an integral component of the patient counselling process.Patient counselling with shared decision-making is crucial in the consideration and planning of extensive pelvic surgery. A growing volume of research has shown that when patients are actively involved in decision-making and prehabilitation (nutritional, physiotherapy/conditioning and/or psychological input) that they experience better outcomes [144,145]. The routine use of patient-reported outcome measures (PROMs) will inform us as to how surgery impacts a patient’s lifestyle and quality of life [146,147]. Standardized questionnaires that collect data on patients post-operatively, particularly regarding symptoms, health-related quality of life and functional status are vital [148]. While there is ample support for the use of PROMs in the literature, there has been limited uptake amongst surgeons [149,150]. PROMs not only inform practitioners of the nature, frequency and impact of adverse events following treatment but may also be used to identify and treat these effects on an individual level in the post-operative period. The PelvEx Collaborative has supported the development of a specific PROM QoL tool via the European Organisation for Research and Treatment of Cancer network. Survivorship is at the forefront of this project, in the hope that it will give greater insight into the most important factors affecting patients and strategies. The morbidity of pelvic exenterative surgery may extend long into the months after discharge from hospital, and it is imperative that we have the necessary supports in place to manage these complications [151]. Post-operative specialist multi-disciplinary care is essential to assist patients with pain, wound and stoma management as well as for psychological support.The role of radical surgery in the setting of locally advanced and recurrent rectal cancer has evolved substantially. Novel strategies including TNT, cytoreductive and/or bony/vascular resection and enhanced reconstructive techniques have enabled surgeons to pursue what was once considered a terminal disease. Advancements in surgical technology, in particular the incorporation of artificial intelligence and three-dimensional bioprinting, will undoubtedly enhance our ability to move the limits of what is reasonable and possible to resect.All authors have been involved in the preparation, reviewed and approval of this manuscript by the PelvEx Collaborative. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The authors declare no conflict of interest.PelvEx Collaborative: Kelly M.E., O’Sullivan N.J., Fahy M.R., Aalbers A.G.J., Abdul Aziz N., Abecasis N., Abraham-Nordling M., Abu Saadeh F., Akiyoshi T., Alberda W., Albert M., Andric M., Angeles M.A., Angenete E., Antoniou A., Auer R., Austin K.K., Aytac E., Aziz O., Bacalbasa N., Baker R.P., Bali M., Baransi S., Baseckas G., Bebington B., Bedford M., Bednarski B.K., Beets G.L., Berg P.L., Bergzoll C., Beynon J., Biondo S., Boyle K., Bordeianou L., Brecelj E., Bremers A.B., Brunner M., Buchwald P., Bui A., Burgess A., Burger J.W.A., Burling D., Burns E., Campain N., Carvalhal S., Castro L., Caycedo-Marulanda A., Ceelen W., Chan K.K.L., Chang G.J., Chang M., Chew M.H., Chok A.Y., Chong P., Clouston H., Codd M., Collins D., Colquhoun A.J., Constantinides J., Corr A., Coscia M., Cosimelli M., Cotsoglou C., Coyne P.E., Croner R.S., Damjanovich L., Daniels I.R., Davies M., Delaney C.P., de Wilt J.H.W., Denost Q., Deutsch C., Dietz D., Domingo S., Dozois E.J., Drozdov E., Duff M., Eglinton T., Enriquez-Navascues J.M., Espín-Basany E., Evans M.D., Eyjólfsdóttir B., Fearnhead N.S., Ferron G., Fichtner-Feigl S., Flatmark K., Fleming F.J., Flor B., Folkesson J., Frizelle F.A., Funder J., Gallego M.A., Gargiulo M., García-Granero E., García-Sabrido J.L., Gargiulo M., Gava V.G., Gentilini L., George M.L., George V., Georgiou P., Ghosh A., Ghouti L., Gil-Moreno A., Giner F., Ginther D.N., Glyn T., Glynn R., Golda T., Griffiths B., Harris D.A., Hanchanale V., Harji D.P., Harris C., Helewa R.M., Hellawell G., Heriot A.G., Hochman D., Hohenberger W., Holm T., Hompes R., Hornung B., Hurton S., Hyun E., Ito M., Iversen L.H., Jenkins J.T., Jourand K., Kaffenberger S., Kandaswamy G.V., Kapur S., Kanemitsu Y., Kazi M., Kelley S.R., Keller D.S., Ketelaers S.H.J., Khan M.S., Kiran R.P., Kim H., Kim H.J., Koh C.E., Kok N.F.M., Kokelaar R., Kontovounisios C., Kose F., Koutra M., Kristensen H.Ø., Kroon H.M., Kumar S., Kusters M., Lago V., Lampe B., Lakkis Z., Larach J.T., Larkin J.O., Larsen S.G., Larson D.W., Law W.L., Lee P.J., Limbert M., Loria A., Lydrup ML., Lyons A., Lynch A.C., Maciel J., Manfredelli S., Mann C., Mantyh C., Mathis K.L., Marques C.F.S., Martinez A., Martling A., Mehigan B.J., Meijerink W.J.H.J., Merchea A., Merkel S., Mehta A.M., Mikalauskas S., McArthur D.R., McCormick J.J., McCormick P., McDermott F.D., McGrath J.S., Malde S., Mirnezami A., Monson J.R.T., Navarro A.S., Neeff H., Negoi I., Neto J.W.M., Ng J.L., Nguyen B., Nielsen M.B., Nieuwenhuijzen G.A.P., Nilsson P.J., Nordkamp S., Nugent T., Oliver A., O’Dwyer S.T., Paarnio K., Palmer G., Pappou E., Park J., Patsouras D., Peacock O., Pellino G., Peterson A.C., Pfeffer F., Pinson J., Poggioli G., Proud D., Quinn M., Quyn A., Rajendran N., Radwan R.W., Rajendran N., Rao C., Rasheed S., Rausa E., Regenbogen S.E., Reims H.M., Renehan A., Rintala J., Rocha R., Rochester M., Rohila J., Rothbarth J., Rottoli M., Roxburgh C., Rutten H.J.T., Safar B., Sagar P.M., Sahai A., Saklani A., Sammour T., Sayyed R., Schizas A.M.P., Schwarzkopf E., Scripcariu D., Scripcariu V., Selvasekar C., Shaikh I., Simpson A., Skeie-Jensen T., Smart N.J., Smart P., Smith J.J., Solbakken A.M., Solomon M.J., Sørensen M.M., Sorrentino L., Steele S.R., Steffens D., Stitzenberg K., Stocchi L., Stylianides N.A., Swartling T., Spasojevic M., Sumrien H., Sutton P.A., Swartking T., Takala H., Tan E.J., Taylor C., Taylor D., Tekin A., Tekkis P.P., Teras J., Thaysen H.V., Thurairaja R., Thorgersen E.B., Tiernan J., Toh E.L., Tolenaar J., Tsarkov P., Tsukada Y., Tsukamoto S., Tuech J.J., Turner W.H., Tuynman J.B., Valente M., van Ramshorst G.H., van Rees J., van Zoggel D., Vasquez-Jimenez W., Vather R., Verhoef C., Vierimaa M., Vizzielli G., Voogt E.L.K., Uehara K., Urrejola G., Wakeman C., Warrier S.K., Wasmuth H.H., Waters P.S., Weber K., Weiser M.R., Wheeler J.M.D., Wild J., Williams A., Wilson M., Wolthuis A., Yano H., Yip B., Yoo R.N., Zappa M.A., Winter D.C.Maximizing success in pelvic exenterative surgery.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01162.txt

@@ -0,0 +1 @@
+Despite recent progress on the treatment of metastatic uveal melanoma (mUM), prognosis remains dismal for the majority of patients. Directed liver therapies including selective internal radiation therapy (SIRT) have been the pillar of hepatic metastases management. Independently, immune checkpoint blockade by combination of ipilimumab plus nivolumab has demonstrated a median survival slightly superior to 1 year. However, the benefit of sequential ipilimumab plus nivolumab immunotherapy and SIRT has not been elucidated.To assess the safety and efficacy of ipilimumab plus nivolumab around selective internal radiation therapy (SIRT) in patients with metastatic uveal melanoma (mUM). We present a retrospective, single center study of 32 patients with mUM divided into two groups based on the treatment received between April 2013 and April 2021. The SIRT_IpiNivo cohort was treated with Yttrium-90 microspheres and ipilimumab plus nivolumab before or after the SIRT (n = 18). The SIRT cohort underwent SIRT but did not receive combined immunotherapy with ipilimumab plus nivolumab (n = 14). Twelve patients (66.7%) of the SIRT_IpiNivo arm received SIRT as first-line treatment and six patients (33.3%) received ipilimumab plus nivolumab prior to SIRT. In the SIRT group, seven patients (50.0%) received single-agent immunotherapy. One patient treated with combined immunotherapy 68 months after the SIRT was included in this group. At the start of ipilimumab plus nivolumab, 94.4% (n = 17) presented hepatic metastases and 72.2% (n = 13) had extra liver disease. Eight patients (44.4%) of the SIRT_IpiNivo group experienced grade 3 or 4 immune related adverse events, mainly colitis and hepatitis. Median overall survival from the diagnosis of metastases was 49.6 months (95% confidence interval (CI); 24.1-not available (NA)) in the SIRT_IpiNivo group compared with 13.6 months (95% CI; 11.5-NA) in the SIRT group (log-rank p-value 0.027). The presence of extra liver metastases at the time of SIRT, largest liver lesion more than 8 cm (M1c) and liver tumor volume negatively impacted the survival. This real-world cohort suggests that a sequential treatment of ipilimumab plus nivolumab and SIRT is a well-tolerated therapeutic approach with promising survival rates.Uveal melanoma (UM) is a rare cancer with an incidence in Europe that varies from <2 per million in Spain and southern Italy up to >8 per million in Northern countries such as Norway or Denmark [1]. However, it is the most frequent primary intraocular malignancy. Despite effective therapy for primary tumors, 50% of patients will develop metastases, principally in the liver [2,3,4]. The prognosis of metastatic disease is poor with survival rates around 20% at 1 year, and 10% at 2 years [3,4,5]. Long-term survival after diagnosis of metastasis is uncommon.Different systemic treatments have been studied including chemotherapy, targeted therapy and immunotherapy but standard of care does not exist for metastatic uveal melanoma (mUM) patients yet. Most chemotherapies presented overall response rates (ORR) under 5% [6]. Single-agent chemotherapy such as fotemustine, dacarbazine, temozolomide and cisplatin, as well as combined regimens, such as dacarbazine–treosulfan or gemcitabine–treosulfan was investigated with disappointing results [6,7,8,9,10]. Regarding targeted therapy, although selumetinib demonstrated activity in a randomized phase 2 trial, selumetinib associated to dacarbazine did not improve progression-free survival (PFS) compared to dacarbazine and placebo in the phase 3 study [11,12]. Additionally, inhibitors targeting the PI3K/AKT/MTOR pathway showed no benefit or demonstrated limited clinical benefit in phase 2 trials [13,14,15]. Recently, a protein kinase C inhibitor showed modest clinical activity in a phase I study [16]. Lately, tebentafusp, a bispecific antibody that redirects T cell lysis of melanoma cells expressing gp100, demonstrated a prolonged overall survival (OS) as first-line therapy for mUM patients with HLA-A*02:01 compared to investigator’s choice (IC), either pembrolizumab, ipilimumab or dacarbazine [17].Although immune checkpoint inhibitors revolutionized the prognosis of cutaneous melanoma, similar outcomes were not reached in mUM. Immunotherapy based on the single CTLA4 and PD1 checkpoint blockade demonstrated limited activity in mUM [18,19]. PFS of patients treated with ipilimumab varied between 2.8 and 3.6 months and OS between 6.8 and 9.6 months in different trials [20,21]. Anti-PD1 and anti-PDL1 therapies such as pembrolizumab, nivolumab or atezolizumab alone also showed disappointing results with ORR of 3.6% and median overall survival (mOS) of 7.6 months [22]. However, the combination of ipilimumab and nivolumab resulted in longer OS than single-agent immunotherapy, surpassing 1 year in different cohorts with ORR that varies from 11.5 up to 18% [23,24,25,26,27].Given that the liver is the most common site of metastasis affecting up to 90% of patients with mUM, local strategies to treat hepatic disease were widely studied [28,29]. Surgery improved OS in cases of complete resection, nevertheless, only a limited number of patients are eligible for surgical intervention due to the presence of multiple lesions or multi lobar involvement [30]. Liver-directed therapies such as radiofrequency ablation, radiotherapy, chemoembolization, immunoembolization, radioembolization, isolated hepatic perfusion and percutaneous hepatic perfusion, was employed to treat metastatic liver disease [10,31]. Selective internal radiation therapy (SIRT) was demonstrated to be effective in mUM patients with less than 25% of tumor burden [32], as salvage therapy [33] and as first-line treatment [34].Studies suggested that radiotherapy and immunotherapy synergize to enhance the efficacy of the treatments [35]. Two case-series described clinical outcomes of 11 and 12 mUM patients treated by SIRT and sequential immune checkpoints inhibitors. Most of the patients in these studies received either CTLA4 or PD1 inhibitors in monotherapy and the mOS described was around 1.5 years [36,37]. Furthermore, a retrospective review showed a larger mOS with SIRT and concurrent single-agent immunotherapy by ipilimumab, pembrolizumab, nivolumab or IL2 compared to SIRT alone (26 versus 9.5 months) [38]. Few data are currently available on the benefits of combined immunotherapy of ipilimumab plus nivolumab before or after a treatment by SIRT. The objective of our study is to analyze the safety and efficacy of SIRT and sequential immunotherapy combination versus SIRT without combined immunotherapy in patients with mUM.We conducted a single center, observational and retrospective study of mUM patients treated by SIRT between April 2013 and April 2021 at Lausanne University Hospital (CHUV), Switzerland. The population included 18 years and older patients with mUM histologically proven by liver biopsy who underwent SIRT treatment(s) in our center. The analysis was conducted in accordance with the Declaration of Helsinki, the Swiss legal requirements and the principles of good clinical practice. Patients signed the Lausanne University Hospital general consent and accepted the use of their data for research purposes or did not explicitly refuse the use of personal data (following Art. 34 HRA). The protocol was approved by the Research Ethics Committee of Canton de Vaud, Switzerland (protocol no. 2019-00448). Authorized qualified personnel of the CHUV Oncology Department retrieved personal and clinical data from electronic patient records. Available imaging was reviewed by a radiologist and a nuclear medicine radiologist. Date of death was obtained from the Swiss Federal Registry for the Persons. Patients were divided in two arms based on the treatment received. The SIRT_IpiNivo group included patients treated with Yttrium-90 microspheres and ipilimumab plus nivolumab before or after the SIRT. The median time between the first SIRT and the start of ipilimumab plus nivolumab was 7.7 months (95% CI; 7.3–13.0 months). The SIRT group included patients who underwent a SIRT but did not receive ipilimumab plus nivolumab. However, in the SIRT group, 8 patients (57.1%) received immunotherapy. Five of these 8 patients started anti-CTLA4 alone after SIRT and 2 received single anti-PD1, 1 before SIRT and 1 after SIRT. One patient treated with combined immunotherapy 68 months after the SIRT was included in this group due to the very long delay between SIRT and the immunotherapy treatment. A total of 32 patients were included in this study, 18 in the SIRT_IpiNivo group and 14 in the SIRT group. Six patients (33.3%) of the SIRT_IpiNivo group received ipilimumab plus nivolumab treatment prior to the SIRT and the SIRT was performed as first-line therapy in the remaining 12 patients (66.7%). All patients presented a progressive disease (PD) before they received the next treatment, either SIRT or ipilimumab plus nivolumab.The systemic therapy consisted of ipilimumab 3 mg/kg combined with nivolumab 1 mg/kg every 3 weeks for a total of 4 doses followed by nivolumab 3 mg/kg or 240 mg flat dose every 2 weeks. In case of grade 3–4 toxicity or progression, the therapy was interrupted.The 90Y microspheres procedure was carried out according to previously published guidelines [39,40]. Patients underwent a simulation angiography to embolize non-target extrahepatic vessels and avoid any unintentional transmission of the microspheres to non-selected organs. Then, Technetium-99m macro aggregated albumin (99mTc-MAA) was injected into the selected hepatic artery to further assess lung or digestive shunting prior to therapy, tumoral volume targeting and dosimetry. Weeks later, the SIRT was performed as planned to one or both lobes. Whole liver could also be treated in more than one session. Patients underwent imaging (whole body Positron Emission Tomography (PET) scan, liver Magnetic Resonance Imaging (MRI) and/or contrast-enhanced Computed Tomography (CT)) before starting the treatment, and then every 3 months. Liver tumor response to the SIRT was assessed by the modified Response Evaluation Criteria in Solid Tumors (mRECIST) and response to the immunotherapy was evaluated by PET Response Criteria In Solid Tumors (PERCIST) version 1.1. To evaluate liver response to the SIRT, target lesions were assessed at 3 and 6 months after the end of treatment. Adverse events related to the treatments were classified following the Common Terminology Criteria of Adverse Events (CTCAE) version 4.0. Complications related to the SIRT were collected until 30 days after the procedure.OS was analyzed from the first treatment, either radioembolization or ipilimumab plus nivolumab, from diagnosis of metastases and from the first SIRT performed until death or last follow-up. Hepatic progression-free survival (hPFS) was calculated from the first SIRT administered to the liver progression after termination of SIRT treatments. To evaluate liver response to SIRT, the last radiological exam performed a maximum of 30 days before SIRT was considered and for ORR to immunotherapy the last radiological exam before combined immunotherapy. PFS post-SIRT was calculated from the first SIRT performed, and PFS post-immunotherapy from the first cycle of ipilimumab plus nivolumab until disease progression. Survival curves were calculated using the Kaplan–Meier method and the hazard ratio (HR) and 95% of CI using a Cox model. Within the R Statistical Computing environment v4.0.3, the packages used are survival, survminer and ggplot2 [41,42]. Significance is defined as a p-value < 0.05 for the log-rank test.Median age at diagnosis of metastases was 61 years in both groups. The median time from primary tumor diagnosis to the development of metastases was 28.8 and 21 months, respectively. Patient and SIRT treatment characteristics are summarized in Table 1.At the time of SIRT, 8 patients (44.4%) in the SIRT_IpiNivo group and 3 patients (21.4%) in the SIRT group presented extra liver metastases. At the start of the immunotherapy combination, 94.4% (n = 17) of individuals of the SIRT_IpiNivo group presented with hepatic disease and 13 patients (72.2%) had extra liver metastases. Eleven patients (61.1%) presented with metastases in 3 or more organs. The most frequent site of extra liver metastases was lung (10 patients, 55.6%). Localization of extrahepatic metastases of the SIRT_IpiNivo group are described in Table 2.Genomic testing was available in 88.9% of cases (n = 16) in the SIRT_IpiNivo group but only in 4 (28.6%) in the SIRT group. Hotspot mutations GNAQ or GNA11 were present in more than 90% of cases. In addition, somatic BAP1 mutation was found in 3 tumors (18.8%) and 1 tumor (6.2%) had a SF3B1 mutation. FGFR1 deletion and FGFR4 mutation were found in 2 different patients (11.1%). In the SIRT group, one tumor of a patient presented with SF3B1 mutation besides GNAQ mutation. Genomic characteristics are shown in Table 1.A total of 52 SIRT treatments were performed, 31 in the group SIRT_IpiNivo and 21 in the group SIRT. The median time from diagnosis of liver metastases to the first SIRT was 3.5 months (range: 1.0–29.8 months) in the SIRT_IpiNivo group and 2.3 months (range: 1.2–11.1 months) in the SIRT group. Median activity infused per patient was 2.4 GBq (range: 1.1–9 GBq) and 2.3 GBq (range: 1.3–5.4 GBq), respectively. SIRT treatment characteristics are shown in Table 1.Concerning the ipilimumab plus nivolumab immunotherapy received by patients in the SIRT_IpiNivo group, it was the first systemic treatment for 16 of 18 patients (88.9%). The 2 remaining patients received sorafenib prior the immunotherapy as part of study protocol SIRT-Sorafenib (NCT01893099). Eleven patients (61.1%) completed 4 cycles of ipilimumab plus nivolumab. Three of these patients did not continue maintenance therapy with nivolumab due to PD. Characteristics of ipilimumab plus nivolumab treatment received by patients of the SIRT_IpiNivo group are described in Table 2.There were no deaths related to treatment. Complications related to SIRT and to ipilimumab plus nivolumab are summarized in Table 3. Immune-related adverse events (irAEs) were described in 12 patients (66.7%) of the SIRT_IpiNivo group. The irAEs most commonly developed were hepatitis and colitis. Eight patients (44.4%) presented with grade 3–4 complications, all of which received corticoids. Four cases needed a second immunosuppressive drug. Three of 7 patients who discontinued ipilimumab plus nivolumab due to toxicity, resumed immunotherapy with single-agent checkpoint nivolumab, as soon as the immune related complication was resolved. These patients discontinued nivolumab later, 2 cases due to immune related complications and the remaining case due to PD.Regarding toxicities related to the SIRT, a grade 3 complication was described in 1 patient of each group. The most frequent complication was abdominal pain. The grade 3 complications were a celiac arterial dissection treated by angioplasty and an enterocolitis managed by intravenous antibiotics. Importantly, 10 patients (55.6%) of the SIRT_IpiNivo group and 6 patients (42.8%) of the SIRT group presented with abnormal hepatic tests before the SIRT was performed. The median follow-up from the diagnosis of metastases was 23.9 months (range: 1.9–91.3 months). The mOS from the diagnosis of metastatic disease was 49.6 (95% CI; 24.1-NA months) in the SIRT_IpiNivo group compared to 13.6 (95% CI; 11.5-NA) in the SIRT group (p-value 0.027), while mOS from the first treatment was 46.6 (95% CI; 22-NA) versus 11.8 (95% CI; 8.5-NA) months (p-value 0.039). The mOS from the first SIRT performed was 46.6 months (95% CI; 18.4-NA) and 11.1 months (95% CI; 8.0-NA), respectively (p-value 0.1) (Figure 1). There was no statistically significant difference in mOS between patients of the SIRT_IpiNivo group who underwent SIRT prior to or after immunotherapy. While the mOS was not reached for SIRT as first line, it was 46.6 months for ipilimumab plus nivolumab prior SIRT. Liver response following SIRT was observed in 9 (50%) and 5 (35.7%) patients of SIRT_IpiNivo and SIRT groups, respectively (Figure 2). Liver response to SIRT was comparable in the SIRT_IpiNivo group, regardless of whether ipilimumab plus nivolumab was administered prior to or after SIRT.Median hPFS and median progression-free survival (mPFS) from the first SIRT was 8.7 (95% CI; 4.4-NA) and 4.6 (95% CI; 2.7–10) months in the SIRT_IpiNivo group and 5.6 (95% CI; 4.9-NA) and 4.9 (95% CI; 4.1-NA) months in the SIRT group (Figure 3). The median hepatic duration of response was 13.3 months (range: 2.1–28.5 months) in the SIRT_IpiNivo group versus 7.9 months (range: 4.9–15.3 months) in the SIRT group. Median hPFS was 8.7 months for patients who received SIRT prior to ipilimumab plus nivolumab and 10.2 months for upfront immunotherapy (no statistically significant difference).At the time of PD following SIRT, 6 patients (33.3%) of the SIRT_IpiNivo group and 8 patients (57.1%) of the SIRT group presented progression of liver and extra liver metastasis; 3 and 2 patients, respectively (16.7% and 14.3%), had progression exclusively in the liver while 9 patients (50.0%) of the SIRT_IpiNivo group and 3 patients (21.4%) of the SIRT group presented only extra liver progression. Responses to SIRT at 3 and 6 months are summarized in Table S1. In the group SIRT_IpiNivo, ORR following ipilimumab plus nivolumab was 22.2% (n = 4). The disease control rate was 38.9% (n = 7). Two patients (11.1%) presented a complete response (CR). One of these two patients presented liver metastases only and the second patient presented with extrahepatic and hepatic metastases. This patient also had 3 millimetric brain metastases controlled by stereotaxic radiotherapy. The median duration of response (DoR) to immunotherapy for the SIRT_IpiNivo group was 22.3 months (range: 8.8–42.2 months). Univariate analysis was performed to identify factors influencing hPFS and OS (Table S2). Patients with an index liver tumor greater than 8 cm (M1c) at the first SIRT had decreased hPFS and OS from SIRT when compared with patients with the largest hepatic lesion smaller than 8 cm (p-value 0.00096; HR: 3.47 (1.19–10.07) for hPFS and p-value 0.00059; HR: 6.75 (1.95–23.35) for OS) (see Figure S1). Tumoral volume superior to the median (185 cc) also had a negative impact on hPFS (p-value 0.016; HR: 2.62 (1.17–5.86)) and OS from SIRT (p-value 0.046; HR: 2.82 (0.98–8.17)). Presence of extra liver metastatic disease at SIRT was prognostic but not predictive (p-value 0.034; HR: 2.82 (1.04–7.66) for OS from SIRT).This study describes the safety and efficacy of sequential combination of ipilimumab plus nivolumab and SIRT in patients with mUM compared with SIRT without combined immunotherapy. Our study showed that both therapies are well tolerated. Ipilimumab plus nivolumab before or after the SIRT was associated with improved OS in our retrospective analysis.We analyzed complications related to SIRT as well as irAEs after immunotherapy combination. The frequency of grade 3 toxicities due to SIRT in our study was comparable to those described in other analyses [36,38]. Nevertheless, liver function test elevation was observed in more than 90% of patients in our study, which is notably higher than rates detailed previously, probably because of pre-existing liver test abnormalities presented by around half of our patients (55.5%). Most complications related to SIRT were treated conservatively. The safety profile of combined checkpoint inhibition of our analysis was also consistent with other cohorts. In our study, 44.4% of the SIRT_IpiNivo group developed grade 3 irAEs, whereas the prospective clinical trials of Piulats et al. and Pelster et al. who analyzed 52 and 35 patients with mUM treated by ipilimumab plus nivolumab, described grade 3 immune side effects in 57.7% and 40% of patients, respectively [26,27]. Najjar et al. described grade 3 immune side effects in 30% of an 89 mUM patient cohort treated with ipilimumab plus nivolumab [25]. The frequency of severe histologically proven grade 3 or 4 hepatitis was higher in our study (n = 4, 22.2%) compared with cutaneous melanoma patients treated with the same regimen. In the study of Larkin et al., 313 patients were treated with ipilimumab plus nivolumab and only 8.3% developed a severe increase in alanine amino-transferase and 6.1% of aspartate amino-transferase levels [43]. It is important to mention that the majority of patients in our study presented liver disease (94.4%) and pre-existing liver function test elevation (72%) at the start of immunotherapy and these conditions could eventually contribute to the hepatotoxicity developed by our patients. The frequency of immune-related hepatitis was greater with ipilimumab and nivolumab after SIRT (33.3%) than with upfront immunotherapy (16.7%).As the sequence of treatments explored in our study was not homogenous, for instance a third of patients of the SIRT_IpiNivo group received immunotherapy prior SIRT and the remaining two thirds underwent SIRT before immunotherapy, to evaluate the benefit from the immunotherapy as well as the SIRT, we calculated mOS from the diagnosis of metastases and from the administration of the first treatment of either combined immunotherapy or SIRT. In both cases, mOS was significantly superior in the SIRT_IpiNivo group compared to the SIRT group. In addition, half of the SIRT group had previously received ipilimumab. However, as ORR to ipilimumab described in previous studies was extremely low, we did not expect a significant impact on the outcomes of this study. Additionally, a significant negative impact on survival rates from SIRT in univariate analysis were observed with the presence of extra liver metastases at the time of SIRT, larger liver lesions (M1c) and higher tumor volume (Table S2). The variables that reflect liver tumor burden also affected negatively the hPFS in our analysis, suggesting that a SIRT treatment should not be delayed in the presence of unresectable liver metastases. Patients with index liver metastasis greater than 8.0 cm (M1c according to AJCC 8th edition staging system) had negative repercussions in hPFS and OS from SIRT compared to those with smaller lesions. Levey et al. already described this effect when the largest liver lesion was greater than 7 cm [38].Sequential treatment of SIRT and immunotherapy has been previously analyzed in retrospective studies [35,36,37]. Given that UM is a rare disease, these reports are limited by small sample size as it is the case of our study. Levey et al. analyzed 24 mUM patients treated by SIRT and concluded that the subgroup of 12 patients treated consecutively by SIRT and immunotherapy (ipilimumab, nivolumab, pembrolizumab or IL2) within 3 months before or after the SIRT, presented longer survival rates with regard to SIRT alone [38]. Ten patients of this review received the immunotherapy treatment prior to the SIRT. Ruohoniemi et al. presented a cohort of 22 patients, which included 12 mUM patients treated with radioembolization and immunotherapy with ipilimumab, nivolumab, pembrolizumab or ipilimumab plus nivolumab combination (n = 7) within a 15-month period [36] while Zheng et al. studied 11 patients treated by SIRT and anti-PD1 or anti-CTL4 alone. Nine of these cases received the immunotherapy before SIRT [37]. The mOS from the SIRT of these three studies was 18.6, 20 and 17 months, respectively. Survival rate from the diagnosis of liver disease of patients treated by SIRT and immunotherapy was reported by Zheng et al. and Levey et al. as 35.5 and 26 months, respectively. Additionally, Blomen et al. recently showed in a retrospective study of two cohorts, immunotherapy and liver directed therapy compared to standard therapies, that mOS showed significant improvement in the first cohort (22.5 versus 11 months) [44]. In our study, the mOS of the group treated by ipilimumab plus nivolumab and SIRT was 46.6 months (95% CI; 18.4-NA) from the SIRT and 49.6 (95% CI; 24.1-NA) from the diagnosis of metastases, compared with 11.1 (95% CI; 8.0-NA) and 13.6 (95% CI; 11.5-NA) months for those treated with SIRT without combined immunotherapy. Although a significant statistical difference was noted when the analysis was calculated from the diagnosis of metastasis (p-value = 0.027), the OS benefit with ipilimumab plus nivolumab was not statistically significant when survival was estimated from the SIRT. Whereas Levey et al. included only patients who received immunotherapy within 3 months of undergoing the SIRT, in our study six patients (33.3%) started ipilimumab plus nivolumab between 5.9 and 19.6 months before SIRT. Moreover, we cannot exclude a lead-time bias that would tend to overestimate the OS of the SIRT_IpiNivo group, due to the time interval between the treatments, during which no death can occur by definition. A comparative table recapitulating previous studies can be found in the Supplementary Materials (Table S3).Although single-agent immunotherapy has not been effective in mUM, survival rates have been improved with combined CTLA4 and PD1 antibodies. In the retrospective analysis of Heppt et al. the mOS was 14 months for pembrolizumab, 10 months for nivolumab and it was not reached with combined immunotherapy. Nevertheless, the follow up for this group was only 3.9 months [23]. The largest cohort assessing combined immunotherapy in mUM patients, collected clinical data retrospectively from 89 patients [25]. Despite a modest PFS of 2.7 months, the mOS from treatment initiation was 15 months. Only two prospective studies have reported outcomes of combined ipilimumab plus nivolumab in mUM patients. GEM1402 included 50 treatment-naïve patients and PROSPER enrolled 35 patients accepting previously treated patients. In GEM1402, the mOS surpassed 1 year (12.7 months) despite mPFS being comparable to monotherapy [27]. PROSPER described 18% of ORR, 5.5 months of mPFS and 19.1 months of mOS. ORR and PFS since the 1st cycle of ipilimumab plus nivolumab of our analysis did not differ from those mentioned. While the ORR was 22.2%, mPFS was 4.4 (95% CI; 2.6–6.5) months [26]. However, comparison with previous trials should be done carefully as clinical characteristics differ largely between studies and differences may not be significant with this sample size. For instance, 50.0% of SIRT_IpiNivo patients had a high level of LDH versus 37.2% in GEM1402 and 43% in PROSPER. Furthermore, 8 of 9 patients (88.9%) of the SIRT_IpiNivo arm with BAP1 status available, exhibited a loss of expression of BAP1 associated with an increased risk of metastasis and a poor prognosis according to Robertson et al. [45]. SF3B1 mutated tumors may respond better to checkpoint inhibitors because SF3B1 alternative splicing may generate neo-antigens [46,47,48]. However, SF3B1 mutation is found in only 20%–25% of primary UM and this group has been classified as an intermediate metastatic risk [45]. Therefore, the frequency of SF3B1 mutation in our cohort is expected to be negligible. Regarding the distribution of metastases, whereas 67% of our patients presented with liver and extra liver disease at the first cycle of immunotherapy, 95% had liver involvement. GEM1402 and PROSPER reported hepatic and extrahepatic disease in 37% and 49% of patients, respectively. Moreover, around 20% of patients included in GEM1402 and PROSPER had extra liver metastases exclusively. Patients of GEM1402 with extra liver metastasis had longer survival rates regardless of liver status.Regarding the response to SIRT, a systematic meta-analysis of 55 studies including 2446 patients evaluated different treatments of UM liver metastases and suggested an improvement of survival with surgery and locoregional procedures [29]. The mOS from the SIRT of five clinical studies reported in this meta-analysis varied between 2.9 and 12.3 months [32,33,49,50,51]. The largest cohort of mUM patients treated by SIRT was described by Eldredge-Hindy et al. and included 71 patients [33]. SIRT was administered as a salvage therapy in 82% of cases and the mOS was 12.3 months. When SIRT is administered as first line treatment, the mOS increases to 18 months as reported by Ponti et al. [34]. Tulokas et al. compared patients treated by SIRT to an historical mUM group without extrahepatic spread treated by systemic chemotherapy as first line treatment. The mOS was 13.5 months for patients treated by SIRT, significantly longer than the 10.5 months (p-value 0.047) of the historical group [52]; however, the mOS increased to 18.7 months with SIRT as first line treatment (p-value 0.017). In our cohort, 12 patients of every group received SIRT as the first line. The mOS since the first SIRT estimated for these patients, regardless of the group, was 22 months (95% CI; 10.8–86.1 months), comparable to studies previously described. Nevertheless, the mOS of patients managed with SIRT as the first line was 9.8 months (95% CI; 8.0-NA) for the SIRT group compared to 46.6 months (95% CI; 18.4-NA) in the SIRT_IpiNivo group. Despite a notably longer survival of patients treated with combined immunotherapy in addition to the SIRT, there was no statistically significant difference (p-value = 0.16) between both groups, but the moderate sample size of our cohort could have affected these results. We also reported a mPFS after SIRT comparable to the prospective phase 2 trial of Gonsalves et al. of 48 patients treated by SIRT and pooled into two groups, a naïve treatment group and a group presenting PD after immunoembolization [53]. Nevertheless, in this study, most participants of both groups (100% and 91.3%) developed new hepatic lesions while in our study at the moment of progression on SIRT, 50% of the SIRT_IpiNivo group and 21.4% of SIRT group presented progression of extrahepatic metastases while liver metastases were under control. A high number of our patients presented extra liver metastasis at the time of SIRT, whereas patients with extra liver metastases needing treatment were excluded in the study of Gonsalves et al. It is important to note that although the liver is the most frequent site of metastatic disease in mUM, around 50% will also develop extrahepatic disease [2,28]. It is noteworthy to mention the favorable outcomes recently demonstrated with tebentafusp in a phase 3 trial. The estimated OS at 1 year was 73.2% (95% CI; 66.3–78.9) in the tebentafusp group versus 57.5% (95% CI; 47.0–66.6) in the IC group [17]. OS was superior for patients receiving tebentafusp even when the best response was PD (HR = 0.41, 95% CI; 0.25–0.66) [54]. While tebentafusp is currently the best option for mUM patients, demonstrated in a randomized study, its use is restricted to patients with HLA-A*02:01, found in around 50% of patients with UM [55]. Therefore, investigations to establish optimal treatment sequences for mUM are still needed.The limitations of our study include a small cohort size, a retrospective analysis, and a non-randomized setting. In addition, the baseline characteristics were heterogenous and the order of treatments differed among patients in the SIRT_IpiNivo group. Furthermore, half of the patients in the SIRT group received single-agent immunotherapy, two patients in the SIRT_IpiNivo group had systemic therapy before the immunotherapy and four benefited from liver directed therapy prior to SIRT. Additionally, all patients of this study presented with PD on the first treatment, either SIRT or combined immunotherapy, before receiving the subsequent treatment. Despite these limitations, this is the first report, to our knowledge, comparing combined checkpoint inhibition sequentially with SIRT versus SIRT without ipilimumab plus nivolumab in mUM patients.We conclude that combined immunotherapy before or after the SIRT is a safe therapeutic option and appears to be associated with improved survival rates. There were no statistical differences between patients that received upfront immunotherapy compared with patients that received SIRT as first treatment. The approach needs to be further investigated in prospective studies, in particular to define the best sequence of therapies. The hypotheses based on our real-world data might eventually be confirmed with the results of the ongoing NCT02913417 study that is currently evaluating the safety and tolerability of radioembolization and immunotherapy by ipilimumab plus nivolumab treatment started 3–5 weeks after SIRT. The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051162/s1, Table S1: Best liver response to SIRT. Table S2: Univariate analysis of overall survival and hepatic progression-free survival. Table S3: Comparative table of outcomes of different studies using ipilimumab and nivolumab and/or SIRT as treatment for metastatic uveal melanoma. Figure S1: Survival curves and hepatic progression-free survival from SIRT.Conceptualization and methodology, V.A.-L., M.A.C., R.D. and O.M.; validation V.A.-L. and C.L.G.; formal analysis, V.A.-L., M.A.C., C.L.G.; R.D. and O.M.; project administration, V.A.-L. and C.L.G.; data curation, V.A.-L. and C.L.G.; resources, V.A.-L., C.L.G.; S.B., G.B. and B.G.M.; writing—original draft preparation, V.A.-L.; writing—review and editing, V.A.-L., M.A.C., C.L.G., S.B., B.G.M., G.B., A.D., K.H., N.S., R.D., O.M.; visualization, V.A.-L. and C.L.G.; supervision, O.M. and M.A.C. All authors have read and agreed to the published version of the manuscript.This research received no external funding.This study was conducted in accordance with the Declaration of Helsinki, the Swiss legal requirements, and the principles of Good Clinical Practice. The protocol was approved by the Research Ethics Committee—Vaud Canton, Switzerland (Protocol No. 2019-00448).Patients signed the Lausanne University Hospital general consent and accepted the use of their data for research purposes or did not explicitly refuse the use of personal data (following Art. 34 HRA).All data relevant to the study are included in the article or uploaded as online Supplementary Information (Table S3).V.A.-L, C.L.G., S.B., G.B., K.H., R.D. and N.S. declare no competing interest to report. B.G-M. is a consultant or participated to advisory boards and is beneficiary of research funding from Bristol Myers and Squibb, travel accommodation for Bristol Myers and Squibb, Merck, Sharp & Dohme, Pierre-Fabre, and Novartis. A.D. is a consultant or participated to advisory boards for Servier, Pharmamar, Bayer and Bristol Myers and Squibb. M.A.C. is beneficiary of research funding from Merck Serono. O.M. is the beneficiary of research funding from Merck, Sharp & Dohme, Bristol Myers and Squibb and Amgen and is consultant or participated to advisory boards for Roche, Amgen, GSK, BMS, Merck, Sharp & Dohme, Novartis, and Pierre-Fabre.Kaplan–Meier curves for overall survival (OS) since first treatment (A) and from the diagnosis of metastases (B). Kaplan–Meier curve for progression-free survival (PFS) from SIRT (C) and from ipilimumab plus nivolumab (ipi-nivo) (D).(A) Alluvial plot of best response to first treatment received, either SIRT or ipilimumab plus nivolumab (IpiNivo), and best response to the second treatment. (B) Liver responses to SIRT by group. CR: complete response; PR: partial response; SD: stable disease; PD: progressive disease.Swimmer plot. Follow up of each patient from the diagnosis of metastatic disease until death or end of study, time of start of ipilimumab plus nivolumab and first SIRT, best overall response to ipilimumab plus nivolumab and best liver response to SIRT and moment of progressive disease on ipilimumab plus nivolumab and on SIRT.Summary of clinical characteristics, treatment features and response to SIRT.Characteristics of treatment and responses to ipilimumab and nivolumab combination.Summary of complications.* 12 patients (66.6%) developed irAEs, 7 patients developed one irAEs, 4 patients presented two irAEs and 1 patient had three irAEs.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01163.txt

@@ -0,0 +1 @@
+Epigenetic alterations are critical for tumor onset and development. DNA methylation is one of the most studied pathways concerning various types of cancer. A promising and exciting avenue of research is the discovery of biomarkers of early-stage malignancies for disease prevention and prognostic indicators following cancer treatment by examining the DNA methylation modification of relevant genes implicated in cancer development. We have made significant advances in the study of DNA methylation and thyroid cancer. This study is novel in that it distinguished methylation changes that occurred primarily in the gene body region of the aforementioned hypermethylated or hypomethylated thyroid cancer genes. Our findings imply that exposing whole-genome DNA methylation patterns and gene expression profiles in thyroid cancer provides new insight into the carcinogenesis of thyroid cancer, demonstrating that gene expression mediated by DNA methylation modifications may play a significant role in tumor growth.Thyroid carcinoma is the most prevalent endocrine cancer globally and the primary cause of cancer-related mortality. Epigenetic modifications are progressively being linked to metastasis. This study aimed to examine whole-genome DNA methylation patterns and the gene expression profiles in thyroid cancer tissue samples using a MethylationEPIC BeadChip (850K), RNA sequencing, and a targeted bisulfite sequencing assay. The results of the Illumina Infinium human methylation kit (850K) analyses identified differentially methylated CpG locations (DMPs) and differentially methylated CpG regions (DMRs) encompassing nearly the entire genome with high resolution and depth. Gene ontology and KEGG pathway analyses revealed that the genes associated with DMRs belonged to various domain-specific ontologies, including cell adhesion, molecule binding, and proliferation. The RNA-Seq study found 1627 differentially expressed genes, 1174 of which that were up-regulated and 453 of which that were down-regulated. The targeted bisulfite sequencing assay revealed that CHST2, DPP4, DUSP6, ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2 methylation levels were dramatically lowered in thyroid cancer patients when compared to the controls, but GALNTL6, HTR7, SPOCD1, and GRM5 methylation levels were significantly raised. Our study revealed that the whole-genome DNA methylation patterns and gene expression profiles in thyroid cancer shed new light on the tumorigenesis of thyroid cancer.Thyroid carcinoma is currently considered to be induced by the multi-step process of carcinogenesis, in which cancer cells are formed from thyroid follicular cells (thyroid epithelial cells) via numerous incidences of genomic injury. These injuries primarily occur in oncogenes and anti-oncogenes that promote proliferation or the development of malignant phenotypes, such as the ability to penetrate surrounding tissue or metastasize to distant organs [1,2]. Thyroid carcinomas are a frequently occurring type of endocrine cancer that exhibits varying phenotypes, ranging from mild forms to the most aggressive forms of human cancer. Thyroid carcinomas are classified into several types, including well-differentiated thyroid carcinoma (WDTC), undifferentiated thyroid carcinoma (UTC), poorly differentiated thyroid carcinoma (PDTC), anaplastic thyroid carcinoma (ATC), and medullary thyroid cancer (MTC). Of these, WDTC, UTC, PDTC, and ATC are all derived from thyrocytes; in contrast, MTC is derived from C cells. Furthermore, differentiated thyroid cancer is divided into three primary subtypes: papillary thyroid carcinoma (PTC), follicular thyroid carcinoma (FTC), and Hürthle cell cancer. Differentiated thyroid carcinomas account for ninety-five percent of all thyroid cancers globally. Numerous epidemiological studies have found that the incidence of differentiated thyroid cancer has increased significantly over the last few centuries [3]. The vast majority of thyroid carcinomas present as thyroid nodules, which are detected by the physician during a physical examination or during neck imaging for other disorders. Thyroid nodules can be cancerous in a small percentage of cases. Thyroid nodules identified in the general population have a 5–10% chance of being cancerous, although men and patients at the extremes of age are at a greater risk [4]. A considerable proportion of patients with well-differentiated thyroid carcinoma are treated with a total thyroidectomy, including the excision of the anterior or central compartment lymph nodes, radioactive iodine therapy for abscission of metastases and thyroid remnants, and suppression of TSH with l-thyroxin [5].Deeper knowledge of the molecular pathways underlying the growth of thyroid cancer might be critical for tailoring treatments. Over the last three decades, significant progress has been made in this area [6]. There are several epigenetic mechanisms: DNA methylation, chromatin remodeling, and post-translational histone modifications. These mechanisms have been studied elsewhere [7,8,9,10,11]. DNA methylation is a long-lasting epigenetic modification that has identified in cancer for over three decades [12]. As a gene silencing mechanism, DNA methylation is necessary for the proper development and operation of several structures and cellular processes, including embryogenesis, transcription, X-inactivation, and genomic imprinting [13,14,15,16]. In humans, DNA methylation occurs nearly entirely within CpG dinucleotides, which are underrepresented, i.e., they occur less frequently than estimated based on the GC composition of DNA and are not evenly distributed throughout the genome [17]. The vast majority of the human genome is methylated, with approximately 60–80 percent of CG sites methylated, with the exception of selected CpG-rich sections, known as CpG islands or CG islands (CGIs), that are commonly unmethylated and contain the promoters for approximately 60 percent of all protein-coding genes [18,19,20].Although prior studies have shed light on the association between gene regulation and DNA methylation in the development of thyroid cancer, the overall knowledge base remains extremely limited. Multiple studies have demonstrated that aberrant DNA methylation patterns in cancerous tissues may mute tumor suppressor genes while activating oncogenes via hypermethylation/hypomethylation [21,22]. However, hypermethylation is more frequently reported than hypomethylation in cancers [23]. Some genes, including MLH1 and p16INK4A, are frequently hypermethylated in various malignancies, including thyroid carcinoma. The tumor-specific sodium iodide symporter gene, NIS (also known as SLC5A5), is also expressed [24,25,26,27]. Across all of the different epigenetic alterations, DNA methylation on CpG islands is the most widely researched. It is well known that hypermethylation of CpG islands in a gene’s promoter region suppresses its expression. Furthermore, changes in DNA methylation have been observed to arise in the initial stages of oncogenesis, implying that they could be exploited as a viable biomarker for cancer detection [28,29,30]. Several DNA methylation-based biomarkers have been reported in various cancers, including stomach cancer, prostate cancer, bronchial carcinoma, and bowel cancer [31,32,33,34]. The initial studies on the effects of DNA methylation in thyroid carcinoma were performed using a candidate gene approach that assessed the DNA methylation level of particular gene promoters [35].In this study, our primary goal is to detect local differentially methylated CpG regions (DMRs) between thyroid cancer and normal thyroid tissue groups at a genome-wide level. We found 43,653 significantly differentially methylated CpG positions (DMP), accounting for 6.10% of all possible DMPs, and 236 significantly differentially methylated CpG regions, accounting for 18.96% of all possible differentially methylated regions (Supplementary Figure S1). Gene Ontology and KEGG enrichment analysis of differentially methylated and differentially expressed genes (DEGs) revealed that the genes involved in DNA methylation were significantly enriched in the regulation of the phosphatidylinositol 3-kinase/protein kinase B (PI3K-Akt), human papillomavirus (HPV) infection, and mitogen-activated protein kinase (MAPK) signaling pathways. This indicates that methylation-related genes are highly enriched in malignancy-related pathways. In this investigation, we identified 1627 genes expressed differently in tumor tissue than in the adjacent healthy tissue (Supplementary Table S2). In comparison to the nearby normal tissue, the tumor tissue had 453 genes with downregulated expression and 1174 genes with elevated expression. It is yet to be determined which specific differentially methylated genes (DMG) are implicated in thyroid cancer. The hypermethylated and low-expressed genes (thyroid tumor vs. normal control) were intersected, yielding seven genes. The hypomethylated and high-expressing genes (thyroid tumor vs. normal control) were intersected, yielding 65 genes (Supplementary Table S3). The validation of thyroid tumor-related genes reveals that the methylation levels of CHST2, DPP4, DUSP6, ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2 were significantly lower in thyroid cancer patients when compared to the controls, while the methylation levels of GALNTL6, HTR7, SPOCD1, CDH16 and GRM5 were significantly higher (Supplementary Table S5, Figure S8). The thyroid tumor-related genes have been validated. Differently methylated genes are identified using targeted bisulfite sequencing, and differentially expressed genes are identified using RQ-PCR.Pairs of fresh frozen thyroid carcinoma samples and adjacent normal thyroid carcinoma tissue samples were obtained. Initially, we acquired ten pairs of malignant thyroid cancer tissues and normal thyroid tissues, then we went through a quality check of the specimens. We collected a total of 86 matched samples of pathologically verified post-operative malignant carcinoma and normal thyroid tissues from 43 specimens, from patients who underwent thyroidectomy at the Zhenjiang First People’s Hospital, affiliated with the Institute of Jiangsu University, between July 2018 and September 2020. The overall structure and the methods used in this study are shown in Figure 1. All procedures were conducted following the appropriate norms of Affiliated People’s Hospital of Jiangsu University’s Ethics Board. After surgery, the eighty-six specimens were immediately frozen in liquid nitrogen and preserved at −80 °C. Neither preoperative chemotherapy nor radiotherapy had been administered to the patients selected for this study, and the specimens contained cancerous tissue. The controls were normal thyroid tissues more than 2 cm away from the tumor and did not have infiltrated cancer cells. This study was approved by the Ethics Committee of the Affiliated People’s Hospital of Jiangsu University, and written informed consent was taken from all participants prior to their inclusion.It was determined that the replication cohort’s DNA was methylated using the Infinium MethylationEPIC BeadChip (850K) (Illumina, Inc., San Diego, CA, USA). The R Package ChAMP was used to analyze, normalize, and perform differential methylation analysis on the genome-wide methylation data. Genomic DNA was extracted from cells using the NucleoSpin Tissue kit (Macherery-Nagel, GmbH & Co. KG, Düren, Germany). DNA (cytosine) methylation profiles were generated utilizing an array, by combining bisulfite conversion and whole-genome results amplified with the direct captures and scores of CpG (cytosine-guanine) loci. DNA specimens were processed and hybridized to the human Infinium MethylationEPIC BeadChip (Illumina, San Diego, CA, USA), designed to quantitatively assay over 850K methylation sites across the genome using the Infinium HD Methylation Assay protocol. Hybridized BeadChips were scanned according to the manufacturer’s specifications using an Illumina iScan system. Annotation of the genes was carried out utilizing the annotation provided with Illumina’s probe. To summarize, CpG markers were categorized on the MethylationEPIC 850K array according to their chromosomal position. Marker Infinium (I), Infinium II, and the UCSC annotation feature’s gene area category were studied using Infinium Chemistry [36]. To make the datasets more interpretable, principal component analysis (PCA) reduced their dimensionality while avoiding information loss. This is accomplished by successively increasing the variance of uncorrelated variables. Hierarchical clustering analysis is a comparable technique for grouping similar objects into clusters. The endpoint is a collection of clusters, each distinct from the others, but containing broadly similar objects. Genome-wide gene expression analysis of thyroid carcinoma was performed utilizing second-generation RNA sequencing (RNA-Seq). A commercial company evaluated the RNA-Seq data (BGI, Beijing, China). The log2 ratio of the unfiltered air data to the filtered air data represents the gene expression results. Total RNA was extracted according to the manufacturer’s instructions using a combination of TRIzol™ Reagent and ethanol for precipitation (Tiangen Biotech, Beijing, China). A spectrophotometer and an Agilent 2100 Bioanalyzer were used to evaluate the samples’ integrity and RNA content (Agilent Technologies, Inc., Santa Clara, CA, USA).MethylTarget (Genesky, Shanghai, China) was used to perform targeted bisulfite sequencing. MethylTarget is a next-generation sequencing (NGS) platform. As previously mentioned, DNA extraction and bisulfite conversion were carried out [37,38]. To identify the different probable CpG sites in a panel of samples, we meticulously constructed primers based on the genome coordinates of the regions. A net polymerase chain reaction was utilized to amplify the desired DNA sequence, which was then purified. An Illumina HiSeq 2000 sequencing system was then used to sequence the intended DNA fragments. BS-Seeker2, a widely used tool for evaluating bisulfite sequencing data, was employed in our investigation to map bisulfite-treated reads and detect methylation [39]. To complete our methylation process, we examined the bisulfite conversion rates that corresponded to each sample. We then used only samples with a bisulfite conversion rate of 98 percent. After performing a preliminary analysis, we found that each CpG site receives an average of 96% of the nucleotides, and that the missing rate is about 4%. Following additional filtering, sites with coverage less than 20 and sites with a missing rate higher than 0.20 were eliminated. When more than 30% of the samples tested were missing, those samples were also removed.As previously shown, total RNA isolation and reverse transcription were carried out [40]. RQ-PCR was used to validate the significantly up- or down-regulated genes as identified by RNA-Seq analysis. Real-time quantitative PCR (RQ-PCR) was used to detect the expression of the target/screened genes, as previously explained [41]. The study was performed using an BioRad Opticon 2 qPCR machine (Hercules, CA, USA) using an SYBR Green mix (Tiangen Biotechnology, Beijing, China) and the following cycling program: ten min at 95 °C, forty cycles at 95 °C for 25 s, and one min at 60 °C. The internal control for normalization was glyceraldehyde 3-phosphate dehydrogenase (GAPDH) and every sample was ran three times. A set of primers was used to amplify each cDNA aliquot [42].SPSS version 27.0.1 and Prism GraphPad 9.2.0 were utilized for statistical analysis. The Wilcoxon rank-sum test was used to contrast two groups of continuous variables, while Pearson’s chi-squared test was used to differentiate two groups of categorical data. The correlation analysis was conducted using Spearman’s rank correlation coefficient. The statistically significant value was set at p < 0.05 for all tests and conducted on a two-sided basis [42]. The analysis of Gene Ontology (GO) and KEGG (Kyoto Encyclopedia of Genes and Genomes) enrichment was carried out using the R package clusterProfiler. To identify epigenetic alterations occurring in thyroid carcinoma patients, the genome-wide methylation patterns were explored using the Illumina Infinium Human Methylation Assay (850K) in a total of 10 newly diagnosed thyroid carcinoma patients’ tumors and their adjacent tissues. A mean of 864,187 probes per sample was qualified to detect CpG (p-value < 0.01) (Supplementary Table S1). We identified 43,653 significantly differentially methylated CpG positions (DMP), accounting for 6.10% of all possible DMPs, and 236 significantly differentially methylated CpG regions (DMR) accounting for 18.96% of all possible DMRs. The methylation density plot is shown in Supplementary Figure S1. The PCA and hierarchical cluster analysis (HCA) of normal controls and thyroid carcinoma patients based on CpG sites’ methylation differentiated the thyroid carcinoma patients from the controls (except for the three tumor patients labeled as 21T, 22T, and 11T, Supplementary Figures S2 and S3). For the top 300 most significantly differentially methylated CpG positions, tumors from the thyroid carcinoma patients showed drastically increased DNA methylation levels when compared with their normal tissues (Figure 2). Differentially methylated genes were clustered together into hypermethylated and hypomethylated genes, which is obviously displayed in the heat map comparing tumor patients and normal controls (Figure 3).To better understand the potential functions of the DNA methylation-related genes, we conducted a Gene Ontology functional enrichment analysis and a Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis. For biological processes, the results suggested that genes involved in DNA methylation were considerably enriched in terms of cell communication; though material transport, including cell adhesion molecule binding; metal-ion passive transmembrane transporter activity; Ras guanyl-nucleotide exchange factor activity; and cation-ion/substrate-specific channel activity (Figure 4). The KEGG pathway enrichment analysis is shown in Figure 5. The results showed that the regulation of the phosphatidylinositol 3-kinase/protein kinase B (PI3K-Akt), human papillomavirus (HPV) infection, and mitogen-activated protein kinase (MAPK) signaling pathways were considerably enriched in DNA methylation genes. They also indicated that methylation-related genes are significantly enriched in malignancy-related pathways. A comparison of hypermethylated and hypomethylated genes using the KEGG enrichment analysis is given in (Figure 6).Transcriptome analysis of paired tumors and adjacent normal tissues from 10 patients used RNA-Seq was performed to identify changes in gene expression patterns in thyroid cancer. A total of 19,476 genes were investigated for gene expression through RNA-Seq in the tumor patients and normal controls, among which we identified 1627 genes that were expressed differently in tumor tissue, as compared to nearby healthy tissue in the study (Supplementary Table S2). In comparison to the nearby normal tissue, the tumor tissue had 453 genes with down-regulated expression and 1174 genes with elevated expression. Figure 7 shows a heat map of genes that were significantly up- or down-regulated between tumors and healthy tissues.To identify the candidate DMGs involved in thyroid cancer, we screened genes by the following process: first, in the differential methylation results, p-value < 0.001 and |meth. diff| > 0.25 were thresholds to screen out differentially methylated genes. Then, in the differential expression results from whole transcriptome sequencing, p-value < 0.05 and |log2FoldChange| > 1.5 were used as thresholds to screen out differentially expressed genes. Finally, the hypermethylated and low-expressed genes (thyroid tumor vs. standard control) were intersected, and seven genes were obtained. The hypomethylated and high-expressing genes (thyroid tumor vs. standard control) were intersected, and 65 genes were obtained (Supplementary Table S3).After excluding 17 unqualified genes due to them not containing CpG islands, not being feasibile to design primers for, or according to their expression profile, the other 55 DMGs were selected for further validation to give a robust characterization of the methylation state of the CpG sites. Bisulfite sequencing in an additional 36 patients with thyroid carcinoma and 41 normal controls was performed using MethylTarget (Genesky, Shanghai, China), which is based on a next-generation sequencing (NGS) platform. We analyzed 119 DNA target fragments (100–300 bp) from the 55 DMGs and assessed the CpG site-specific methylation levels within each DNA fragment using bisulfate sequencing (Supplementary Table S4). The methylation levels of CHST2, DPP4, DUSP6, ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2 were markedly decreased in thyroid cancer patients when compared with the controls, while the methylation levels of GALNTL6, HTR7, SPOCD1, CDH16 and GRM5 were significantly increased (Supplementary Table S5, Figure 8). In addition, the real-time quantitative PCR (RQ-PCR) results showed that all of the aforementioned genes presented significant differential expression between the 42 patients with thyroid tumors and the 43 healthy controls. All of these genes showed consistent regulation direction within the samples (Supplementary Table S6).Epigenetic modifications play a vital role in the initiation and progression of tumors. DNA methylation has become one of the most investigated mechanisms related to different cancers. It would be very great and promising to discover biomarkers of the early stages of tumors to prevent diseases and prognostic markers following the treatment of cancers by examining the DNA methylation alterations of exciting genes involved in cancer development. In recent years, many efforts have been made by the scientific community in the field of DNA methylation related to thyroid cancer [43,44,45,46]. There is a strong relationship between genetic alterations and epigenetic aberrations in tumorigenesis [47]. It is widely assumed that genetic changes, particularly alternate DNA methylation levels, can significantly impact abnormal gene expression of critical genes, leading to the development and progression of tumors by facilitating or inhibiting transcription factor binding for transcriptional activity [48]. We utilized techniques to focus on the genome-wide level of DNA methylation as to distinguish the hypermethylated and hypomethylated genes and to study the expression level of these genes on a series of paired tumor and adjacent normal tissue samples from patients who acquired thyroid carcinoma in the local region of our city in China. We then selected a few candidate genes which were hypermethylated or hypomethylated in tumor samples as compared to healthy tissues, and then executed a targeted bisulfite sequencing technique to investigate the methylation status of CpG sites in the interesting genes’ promoter regions, as well as their adjacent genomic regions, therefore producing a robust measure of the methylation level of the candidate genes.Several hypermethylated genes (GALNTL6, HTR7, SPOCD1, CDH16 and GRM5) related to thyroid cancer have been discovered to be up-regulated in our study, as were investigated in other reports to mentioned in the following text. GRM5 is a metabotropic glutamate receptor gene that encodes a protein from the G-protein coupled receptor 3 protein family. Its signaling causes a phosphatidylinositol-calcium secondary messenger system to be activated. GRM5 was highly expressed in oral squamous cell carcinoma and contributed to tumor cell migration and invasion [49]. Inhibiting GRM5 expression could suppress oncogenic actions by blocking downstream signaling factors in hepatocellular carcinoma [50]. HTR7 also belongs to the G protein-coupled receptor (GPCR) family. One recent study demonstrated high expression in laryngeal cancer tissues promoted tumor proliferation by activating the PI3K/AKT pathway [51]. SPOCD1 was up-regulated and promoted cell proliferation in osteosarcoma cell lines [52]. It has also been reported that the significantly expressed SPOCD1 accelerated the progression of ovarian carcinoma and inhibited cell death (apoptosis) via the PI3K/AKT pathway [53]. There are currently no reports regarding the high expression of GALNTL6 in cancers. The role of the up-regulation of GALNTL6 in thyroid cancer is still unknown and is worthy of investigation. We have also found a few hypomethylated genes, CHST2, DPP4, DUSP6, ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2, which showed high expression in thyroid carcinoma. Transfer of a sulfate residue to GlcNAc residues in keratan sulfate by CHST2 has been revealed to activate the p38 MAPK-PI3K (mitogen-activated protein kinase/phosphatidylinositol 3-kinase) cell signaling pathway and decrease cell apoptosis caused by radiation in Burkitt’s lymphoma cells [54]. Aberrantly high levels of DPP4 expression occurred in human hepatocellular carcinoma [55]. For the gene DUSP6, a previous study [56] showed that its gene expression increased in all of studied thyroid cancer cell lines, consistent with our results. It also reported that upregulation of DUSP6 gene transcription in human glioblastoma played a tumor-promoting role and accelerated the malignancy of tumors [57]. For ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2, several studies showed that their high expression played a significant role in promoting tumor cell growth and inhibiting cell apoptosis from chemotherapy in different kinds of cancers [58,59,60,61,62]. The most intriguing discovery in our study is that almost all the significant differential methylation alterations occurred exclusively in the region of the gene body of the aforementioned hypermethylated/hypomethylated genes involved in thyroid carcinoma. Gene body methylation is positively correlated with gene expression levels, though the mechanisms are unclear [63]. One recent study revealed that the investigated genes possessed an exceptional hypermethylation level of the CpG islands located in the gene body region, and all of them were simultaneously overexpressed in hepatocellular carcinoma [64]. This phenomenon was believed to be predictive of increased oncogene levels in cancer. Further study needs to be performed to focus on the mechanism behind the methylation alterations in the gene body of these interesting genes in thyroid carcinoma. The results from our study could give potential opportunities to identify new drug targets in thyroid cancer [65].In conclusion, our integrative analysis provides a new perspective that gene expression regulated by DNA methylation alterations located primarily in the gene body may play a crucial role in the progression of tumors and that DNA methylation levels of critical genes could be reverted to normal by methylation or demethylation drugs for the treatment of cancers. Furthermore, these discovered genes can be potentially used as biomarkers for predicting the development of thyroid cancer.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051163/s1, Figure S1: The density distribution plot of the methylation degree of each sample probe after normalization (Note: The X-axis is the degree of methylation, and the Y-axis is the probe density corresponding to the degree of methylation. Each curve represents a sample, and the color distinguishes the sample group); Figure S2: PCA analysis results of methylation ratio of each sample (Note: Color distinguishes sample groups; X and Y axes respectively correspond to the indicators that best reflect the true species composition of the sample, calculated by software. Number represents patients label number, and the following letter N represents patients’ normal control while letter T represents patients’ thyroid carcinoma); Figure S3: Hierarchical clustering results of methylation ratio of each sample. Number represents patients label number, and the following letter N represents patients’ normal control while letter T represents patients’ thyroid carcinoma. Table S1: The number of qualified probes of the sample and the average intensity of the red and green signals of all probes (Note: A maximum of 10 samples are displayed. “Sample Name” refers to the name of each sample; “Detected CpG (0.01)” refers to the number of qualified probes for the sample under the condition that the pvalue threshold is 0.01, that is, the pvalue of the retained probe in all samples is less than 0.01, “Detected CpG (0.05)” refers to the number of qualified probes for the sample under the condition that the pvalue threshold is 0.05, that is, the pvalue of the retained probes in all samples is less than 0.05. “Signal Average GRN” refers to all the samples in each sample. The average intensity of the green signal of the probe, “Signal Average RED” refers to the average intensity of the red signal of all probes for each sample); Table S2: The differential expression of genes between thyroid carcinoma patients and their normal control tissues; Table S3: The screening results of candidate differentially methylated genes related to thyroid cancer; Table S4: The differential methylation level of all target fragments of interested genes by targeted bisulfite sequencing in thyroid cancer; Table S5: The differential methylation level of interested genes by targeted bisulfite sequencing in thyroid cancer; Table S6: the quantitative RT-PCR results showing expression level of targeted genes involved in thyroid cancer.M.A.I.: methodology, investigation, writing—original draft; M.L.: conceptualization, resources, project administration, supervision; J.L.: formal analysis, writing—review and editing; G.Z.: Software; M.C.: visualization, writing—review; N.F.M.: software; W.Q.: data curation. All authors have read and agreed to the published version of the manuscript.This work was sponsored by the Jiangsu provincial “Innovative and entrepreneurial talent team” program.The study was conducted according to the guidelines of the Declaration of Zhenjiang First People’s Hospital, affiliated with the Institute of Jiangsu University, and approved by the Institutional Review Board (or Ethics Committee) of Affiliated People’s Hospital of Jiangsu University’s (K-20210169-W and 2018-01-30).Informed consent was obtained from all subjects involved in the study.No new data were created or analyzed in this study. Data sharing is not applicable to this article.We would like to thank the anonymous reviewers and the editor for their valuable comments and remarks that helped us to improve the original manuscript. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.The overall structure and the methods used for methylation analysis in the thyroid cancer samples.Box plots comparing the degree of methylation of grouped samples for the top 300 most significantly differentially methylated CpG positions. The title indicates the name of the probe, the X-axis indicates the grouping information, the Y-axis indicates the degree of methylation of each grouped sample, and in the upper right corner is the p-value of the probe, as determined by further analysis. The scatter plot shows each sample’s specific methylation degree value, and the box plot shows the difference in the methylation degree distribution between the two groups. The green dots represent thyroid carcinoma from patients, while the red dots represent their adjacent normal tissues.Heat map results of genes where differential methylation sites are located. Each column represents a sample, and each row represents a gene where a differentially methylated site is located. The value of the sample at a given site = log2 (degree of methylation/(degree of 1-methylation)).GO enrichment analysis results of genes with significant methylation differences.KEGG enrichment analysis results of genes with significant methylation differences.GO and KEGG enrichment analysis results of genes with significant methylation differences. (A) GO enrichment analysis of hypermethylated and hypomethylated genes. (B) KEGG enrichment analysis of hypermethylated and hypomethylated genes.Heat map results of genes that are differentially expressed between patients and normal controls. Each column represents a sample, and each row represents a differentially expressed gene.Box plot of the average methylation levels of interesting genes. The name of the image is the name of the target gene. The colors of the dots represent the different groups. The X-axis is the two groups for further analysis. The Y-axis is the average degree of methylation of each sample in the group on the target gene. The box represents the group. Each dot represents each sample’s average degree of methylation on the target gene. (A) CHST2, DPP4, DUSP6, ITGA2, SLC1A5, TIAM1, TNIK, and ABTB2 are hypomethylated in thyroid carcinoma patients when compared to normal control tissues. (B) GALNTL6, HTR7, SPOCD1, CDH16 and GRM5 are hypermethylated in thyroid carcinoma patients when compared to normal control tissues. *: p < 0.05, **: p < 0.01, ***: p < 0.001, ****: p < 0.0001.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01164.txt

@@ -0,0 +1 @@
+Genetically heterogeneous disorder acute myeloid leukemia (AML) is marked by recurring mutations in FLT3. Current FLT3 inhibitors and other emerging inhibitors have helped in the improvement of the quality of standard of care therapies; however, the overall survival of the patients remains static. This is due to numerous mutations in FLT3, which causes resistance against these FLT3 inhibitors. For effective treatment of AML patients, alternative approaches are required to overcome this resistance. Here, we will summarize the biomarkers for FLT3 inhibitors in AML, as well as the alternative measures to overcome resistance to the current therapies.Acute myeloid leukemia is a disease characterized by uncontrolled proliferation of clonal myeloid blast cells that are incapable of maturation to leukocytes. AML is the most common leukemia in adults and remains a highly fatal disease with a five-year survival rate of 24%. More than 50% of AML patients have mutations in the FLT3 gene, rendering FLT3 an attractive target for small-molecule inhibition. Currently, there are several FLT3 inhibitors in the clinic, and others remain in clinical trials. However, these inhibitors face challenges due to lack of efficacy against several FLT3 mutants. Therefore, the identification of biomarkers is vital to stratify AML patients and target AML patient population with a particular FLT3 mutation. Additionally, there is an unmet need to identify alternative approaches to combat the resistance to FLT3 inhibitors. Here, we summarize the current knowledge on the utilization of diagnostic, prognostic, predictive, and pharmacodynamic biomarkers for FLT3-mutated AML. The resistance mechanisms to various FLT3 inhibitors and alternative approaches to combat this resistance are also discussed and presented.Acute myeloid leukemia (AML) is a hematological malignancy that accounts for the most common leukemia that occurs in adults. In the US, there were 20,240 cases and 11,400 deaths due to AML in 2021. The five-year relative survival rate for AML patients is 29.5% [1]. The incidence of AML increases with age, whereby there are 1.3 cases per 100,000 of the population who are under 65 years old, whereas there are 12.2 cases per 100,000 of the population who are 65 years old and above [2]. Several treatment options have improved the survival of younger patients, but the mortality remains high for elderly patients [2,3].AML is characterized by clonal proliferation of poorly differentiated cells of hematologic origin. These cells are genetically altered with recurrent deletions, amplifications, point mutations, and rearrangements [4,5].The human flt3 (FMS-like tyrosine kinase 3) gene is located on chromosome 13q12 and has 24 exons [6]. It encodes a membrane-bound glycosylated protein with a molecular weight of 160 kDa, along with a non-glycosylated isoform which is 143 kDa and not associated with the plasma membrane [7]. FLT3 is a transmembrane protein that encodes for proto-oncogene FLT3. It is a member of the class III receptor tyrosine kinase family and plays an important role in the regulation of the hematopoiesis [8]. The structure of FLT3 consists of four regions: (i) an N-terminal, extracellular region consisting of five immunoglobulin domains involved in ligand binding; the proximal domain is involved in receptor dimerization, (ii) a transmembrane domain, (iii) a juxta membrane domain (JM), and (iv) an intracellular, C-terminal region with a split-kinase domain. The two substructures of this domain are called N-lobe and C-lobe, which are connected by an inter-kinase domain. These lobes consist of a TKD and are also indicated as the first tyrosine kinase (TK1) and second tyrosine kinase (TK2) domain, respectively [9] (Figure 1).The extracellular region contains a binding domain with a high affinity for its ligand (FLT3 ligand or FL). FL is expressed by most tissues, including the spleen, thymus, and bone marrow; however, the highest expression is seen in peripheral blood mononuclear cells [7].Once the FL binds to FLT3, it induces receptor dimerization and conformational changes. Subsequently, FLT3 autophosphorylation activates intracellular signaling cascades that control cell proliferation, differentiation, and survival [7,9,10]. The kinase activity of the FLT3 receptor is negatively modulated by tyrosine phosphatase that dephosphorylates the JM domain. Thus, the frequency of FLT3 production, its degradation, and the downstream effects are regulated by a complex feedback loop for the normal activity of the receptor [7].The most common mutations in FLT3 include FLT3 internal tandem duplication (FLT3-ITDmut), which is detected in 25% of patients, and point mutations in the tyrosine kinase domain (FLT3-TKDmut) that are detected in 7–10% of the patients [11]. These mutations result in the overexpression or constitutive activation of the tyrosine kinase receptor and the downstream proliferative signaling pathways. In addition, FLT3-ITDmut potently activates STAT5, which activates cyclin D1, c-myc, and Pim-2; the activation of these proteins results in the accelerated growth of leukemic cells [7,12]. FLT3-TKDmut consists mainly of missense point mutations, deletions, or insertions in the tyrosine kinase domain of FLT3. The most frequent point mutations are primarily seen in the activation loop in amino acid residues D835, I836, Y842, and some in the TKD1, including the residues N676 and F691 [13].A biomarker is a characteristic that is a measurable indicator of a biological process or response to an intervention. Molecular biomarkers are valuable in providing information about the biological behavior of the AML. These biomarkers can be classified into various categories, including diagnostic, predictive, prognostic, and pharmacodynamic biomarkers, based on their putative applications [14].Diagnostic biomarkers are used to confirm the presence of a disease and aid in the identification of individuals with a disease subtype. These biomarkers are used to identify people with a disease [14]. For example, in the case of AML, gene rearrangements, gene fusions, and chromosomal translocations are used in the diagnosis [15].Predictive biomarkers are used to identify the likelihood of response or lack of response to a particular therapy. These biomarkers help in the identification of patients most likely to benefit from a given treatment and spare other patients from the toxicities of ineffective therapies [14]. NPM1 mutations and the FLT3-ITD allelic ratio (AR) are candidate predictive biomarkers in FLT3 AML [16].Prognostic biomarkers are used to identify the likelihood of a clinical event, disease recurrence, or progression in patients who have the disease [14]. Mutations in the FLT3 gene, such as FLT3-ITD, confer a poor prognosis in AML patients [11]. Pharmacodynamic biomarkers depict the biological response to a medical product or environmental agent in an individual. Such biomarkers are useful for clinical practice and therapeutic development [14]. Various molecular markers, such as phosphorylation, and immune markers have been used in various studies [17,18].All these biomarkers are important because of their high clinical importance, and their expression can reveal the disease evolution in real time [19]. So far, several biomarkers have been identified by various studies and clinical trials of FLT3 inhibitors, which are discussed in the sections below.Various diagnostics have been developed for detecting AML, including morphological, immunophenotyping, and gene fusion screening [20]. For morphological diagnostics, bone marrow smears are examined for myeloblasts, monoblasts, and megakaryoblasts in the blast cells using Wright-Giemsa stains. Immunophenotyping using flow cytometry is used to determine the lineage of leukemia cells [21]. In AML patients, leukemic cells express early, hematopoiesis-associated antigens (CD34, CD38, CD117, HLA-DR) and lack markers of myeloid and monocytic maturation (NSE, CD11c, CD14, CD64) [15,22,23,24]. Similarly, cytogenetic abnormality can be detected in 50% to 60% of newly diagnosed AML patients. The majority of AML patients have nonrandom chromosomal translocations that often lead to gene rearrangements. [25,26,27]. The World Health Organization (WHO) recognizes recurrent translocations and inversions in AML [28,29] (Table 1). Gene rearrangements, gene fusions, and loss of chromosomes are detected using fluorescence in situ hybridization (FISH) and reverse transcriptase–polymerase chain reaction (RT-PCR) [30,31]. These include gene fusions in RUNX1-RUNX1T1 (runt-related transcription factor 1), CBFB-MYH11 (core-binding factor subunit beta–myosin heavy chain 11), acute promyelocytic leukemia (APL) with PML-RARA (promyelocytic leukemia/retinoic acid receptor alpha), MLLT3-KMT2A (mixed-lineage leukemia translocated to chromosome 3- lysine methyltransferase 2A), DEK-NUP214 (DEK oncogene–nucleoporin 214), and an inversion that repositions a distal GATA2 enhancer to activate MECOM expression. BCR-ABL1 is added to recognize that these cases may benefit from tyrosine kinase inhibitor therapy [28,29,32,33]. Finally, for AML diagnosis, testing for mutations in three genes—FLT3, NPM1 (nucleophosmin 1), and CEBPA (CCAAT/enhancer binding protein (C/EBP) alpha)—is recommended [34,35,36]. Additional genes with varying gene mutation frequency in AML patients include mixed-lineage leukemia (MLL), neuroblastoma RAS (NRAS), Wilms’ tumor type 1 (WT1), v-KIT, runt-related transcription factor (RUNX1), and iso-citrate dehydrogenase (IDH1) [37,38,39,40,41,42].Interestingly, recent studies indicated that circulating micro RNAs (miRNAs) can be utilized as diagnostic biomarkers for AML. A study identified six serum miRNAs (miR-10a-5p, miR-93-5p, miR-129-5p, miR-155-5p, miR-181b-5p, and miR-320d) which were specifically upregulated in the serum of AML patients using a next-generation sequencing approach [43,44].Four miRNAs (let-7b, miR-128a, miR-128b, and miR-223) were used for the diagnosis of AML with 97% accuracy and analyzed using RT-PCR. miR-142-3p and miR-29a can also be used as diagnostic biomarkers for AML [45]. Interestingly, miR-424 was downregulated in AML patients with NPM1 mutation regardless of FLT3 mutation, whereas miR-155 was upregulated in patients with FLT3-ITD regardless of the NPM1 mutation [46].These studies suggest that miRNAs from serum or blood samples can be effective diagnostic biomarkers for AML patients.There have been multiple FLT3 inhibitors in clinical trials, but predictive biomarkers remain undiscovered. In a recent study aimed at identifying gene expression changes associated with FLT3 mutation in AML patients, the transcriptomic patterns of six different cohorts of AML patients were analyzed, and a FLT3-mutation-like pattern was highly enriched in NPM1 and DNMT3A mutants. In addition, FLT3-like patterns consisted of numerous homeobox (HOX) genes [47].Based on the FLT3 mutations, companion diagnostics were generated that tested for a predictive biomarker [48]. These tests classified patients into responders and non-responders and directly equated them to the administration of a drug [49].One such FDA-approved companion diagnostic test was the LeukoStrat CDx FLT3 mutation assay. This is a PCR-based, in vitro diagnostic test that detects ITD and TKD mutations D835 and I836 from the genomic DNA extracted from the mononuclear cells from peripheral blood or bone marrow aspirates of AML-diagnosed patients [50].This test was used with FLT3 inhibitors, including midostaurin, gilteritinib, and quizartinib [51].Similarly, co-occurrence of mutations in FLT3 with national comprehensive cancer network (NCCN)-listed gene mutations were used as predictive biomarkers [52,53]. Co-occurrence of mutations in monoallelic, CCAAT/enhancer-binding protein alpha (moCEBPA) with FLT3-ITD/TKD led to a poor prognosis. Mutations in NPM1, DNMT3A, and FLT3-ITD were identified at higher rates in patients with intermediate-risk cytogenetics [54,55]. It was seen that a group of AML patients with FLT3 plus NPM1 and/or DNMT3A mutations shared a similar transcriptomic background [47]. The revised 2017 WHO classification has myeloid neoplasms with germline mutations in RUNX1, CEBPA, DDX41 (DEAD-box helicase 41), RUNX1, GATA2 (GATA binding protein 2), ETV6 (ETS variant transcription factor 6), SRP72 (signal recognition particle 72), and ANKRD26 (ankyrin repeat domain 26) as markers of AML predisposition [29,56,57].Another study identified that the response to gilteritinib and crenolanib among relapsed FLT3mut AML patients is higher in patients with mutations in NPM1 or DNMT3A and particularly in those with both genes mutated [58,59]. When FLT3-ITD leukemias with mutations in NPM1 or DNMT3A are treated with quizartinib, the cell differentiation effect predominates over the cytotoxic mechanism [60]. Additionally, a long non-coding RNA (lnc RNA) expression profile using RNA-seq identified that lncRNA RP11-342 M1.7, lncRNA CES1P1, and lncRNA AC008753.6 serve as predictive biomarkers for AML risk [61].FLT3 is widely overexpressed and the most frequently mutated gene in both pediatric and adult patients with AML [62]. Higher expression of FLT3 results in poor overall survival (OS) in AML patients, as seen in the cancer genome atlas (TCGA) dataset analyzed by GEPIA. The hazard ratio is 1.8 for high-FLT3-expressing patients, indicating that these patients have a ~2 times greater chance of dying compared to the low-FLT3-expressing AML patients [63] (Figure 2).FLT3 ligand (FL) is detectable during homeostasis and is increased in hypoplasia. FL is markedly elevated upon the depletion of the hematopoietic stem or progenitor cells. However, in FLT3+ AML, the levels of FL fall to undetectable levels. It was observed that, after the induction of chemotherapy, FL levels are restored in patients with complete remission but not in patients with refractory disease. FL levels were measured in a randomized study with lestaurtinib where it was seen that patients achieving complete remission (CR) had higher FL levels after the completion of the therapy followed by a normal range after recovery. However, patients with refractory disease had a transient increase in FL levels followed by rapid depletion [64]. Thus, FL levels have the potential to emerge as prognostic biomarkers to guide clinical decisions.The presence or absence of specific gene mutations can be utilized to classify AML patients and determine their prognosis. The NCCN AML prognostic stratification system listed FLT3, NPM1, CEBPA, IDH1/2, DNMT3A (DNA methyltransferase 3A), KIT, TP53 (tumor suppressor 53), RUNX1, and ASXL1 (ASXL transcription factor) gene mutations for the classification of the AML patient population [65,66]. Mutations of NRAS and IDH2 occur in FLT3-independent clones, but TET2 and IDH1 co-occur in FLT3-mutant clones [67].Mutations in the FLT3 gene are of prognostic value for detecting AML in patients. The most common FLT3 mutations (FLT3mut) occur in the JM domain internal tandem duplications, FLT3-ITDmut, or in the tyrosine kinase domain, FLT3-TKDmut. FLT3-ITDmut, are in-frame mutations consisting of duplications of 3–400 base pairs which lead to an elongated JM. This results in constitutive activation of the FLT3 receptor and the downstream signaling (Figure 3) [68,69].The prognostic value of FLT3-ITD is determined by various factors, including the allele ratio (AR), ITD length, karyotype, insertion site, and co-mutations (NPM1) [11,70]. AR is the ratio of ITD-mutated alleles to wild-type alleles (FLT3-ITD/FLT3 wild-type). Similarly, variant allele frequency is determined by the ratio of ITD-mutated alleles to ITD-mutated and wild-type alleles. The European Leukemia Net (ELN) identified a value of 0.5 as a cut-off to distinguish between low and high AR [71]. FLT3-ITD insertion (AR > 0.51) is associated with an unfavorable, relapse-free survival, RFS (p = 0.0008) and OS (p = 0.004) [72]. However, a recent study depicted that the size of FLT3-ITD mutations has no prognostic impact on the overall survival, relapse, or complete remission rate among newly diagnosed AML patients treated with chemotherapy [73].Favorable relapse risk and OS was seen with the occurrence of co-mutations NPM1, along with FLT3mut, in young adult AML patients [74]. In patients with concurrent NPM1mut, the OS and relapse risk were comparable between FLT3 wild-type and FLT3-ITDmut (AR < 0.5), but worse when AR ≥ 0.5 [75]. Among patients with NPM1 wild-type, all FLT3-ITDmut patients had an increased risk of relapse and inferior OS, regardless of the AR. The European Leukemia Net (ELN) guidelines categorize FLT3-ITDmut AML into three categories: favorable (NPM1mut with FLT3 wild-type or NPM1mut with FLT3-ITD AR < 0.5), intermediate (NPM1mut with FLT3-ITD AR > 0.5 or NPM1WT with FLT3-ITD AR < 0.5), and adverse (NPM1WT with FLT3-ITD AR > 0.5) [76]. Although the AR ratio is predictive of the severity of AML in the patients, a strict threshold cannot be established for clinical decision making. This is because the current assays are not optimized, and there is a high intrasample variability [77].FLT3-TKD mutations have prognostic value in the overall AML patient population, but the impact of FLT3-TKDmut AR remains obscure. However FLT3-TKDmut has a high incidence in co-occurrence with mutations in NPM1, CEBPA, and NRAS [78]. These TKD mutations can be identified and detected using next-generation sequencing (NGS). Additionally, computational, biology-based algorithms, such as Pindel, show high sensitivity and specificity in detecting these gene alterations [79].Since FLT3 mutations lead to dysregulation of cell proliferation pathways, inhibiting FLT3 signaling using small molecule inhibitors is a viable strategy for AML patients [80].FLT3 inhibitors can be classified into type I and type II based on their mechanism of interaction with FLT3 (Table 2). Type I inhibitors bind to the gatekeeper domain near the activation loop or the ATP binding site on the receptor regardless of its conformation; however, type II inhibitors bind to the hydrophobic region adjacent to the ATP binding site on the receptor in its inactive conformation. As a result, type I inhibitors can inhibit FLT3 with both ITD and TKD, but type II inhibitors can only inhibit FLT3 with ITD and not TKD [51]. Type I inhibitors include FN-1501, sunitinib, lestaurtinib, midostaurin, crenolanib, and gilteritinib, while type II inhibitors include sorafenib, quizartinib, ponatinib, and pexidartinib [81,82,83,84].FLT3 inhibitors can also be classified into first generation and second generation based on their specificity for FLT3 (Table 2). First-generation inhibitors lack specificity for FLT3. They can bind to multiple receptor tyrosine kinases (RTKs) and inhibit several targets downstream of the FLT3 signaling pathway and parallel pathways, thus, providing a broad range of efficacy in AML patients. Second-generation inhibitors are more specific and only target FLT3. As a result, they are expected to have fewer off-target effects and toxicities. First-generation inhibitors include sunitinib, sorafenib, midostaurin, and lestaurtinib. Some second-generation inhibitors include quizartinib, crenolanib, and gilteritinib [11,96,97].Currently, there are only three FDA-approved FLT3 inhibitors, sorafenib, midostaurin, and gilteritinib, for use in the U.S. Of these, only two are approved for AML indication: midostaurin, along with chemotherapy, and gilteritinib [98]. Midostaurin is a multi-targeted kinase inhibitor with activity against both FLT3-ITD and FLT3-TKD, along with induction chemotherapy with cytarabine and daunorubicin; however, it has limited clinical efficacy as a single agent [87,99]. On the other hand, gilteritinib was recently approved by the FDA with activity against FLT3-ITD, FLT3-TKD, and FLT3, non-canonical mutations in relapsed and refractory (R/R), FLT3-mutated AML patients as a monotherapy [100]. Sorafenib is not approved for the treatment of AML, but off-label use in 13 patients showed improved clinical outcomes in FLT3-ITDmut AML patients [101].Several other FLT3 inhibitors are in the early and late stages of clinical development. Quizartinib is the most potent and selective type II inhibitor and crenolanib is a potent type I inhibitor in the late stages of clinical development (Table 2).Biomarkers that predict the efficacy of FLT3 inhibitors have important applications in clinical care. Activation of FLT3 triggers the phosphorylation and activation of downstream signal transduction pathways including PI3K/AKT/mTOR, RAS/RAF/MAPK and JAK/STAT (Figure 3). Autophosphorylation of the FLT3 receptor has proven to be an excellent biomarker for its activation, and loss of this autophosphorylation is an indication of successful inhibition. The degree of phosphorylation can be quantified directly in the circulating blast cells or the plasma from the patients. Phospho-FLT3 levels can be measured using an enzyme-linked immunosorbent assay (ELISA) and plasma inhibitory activity (PIA) assay. Decreased phosphorylation of FLT3 is associated with clinical activity in patients administered with gilteritinib, midostaurin, and lestaurtinib [18,89,102]. Recently, SEL24/MEN1703, a dual PIM/FLT3 kinase inhibitor, underwent clinical trials for AML patients. This study tested phospho-inhibition of S6, 4-EBP1, and STAT5 as their phosphorylation levels are controlled by both PIM1/2 and FLT3. Preclinical studies identified that S6 phosphorylation (pS6) was at its maximum 4 h post drug treatment; hence, pS6 was chosen as a biomarker for this dual kinase inhibitor (Table 3). pS6 was measured from the whole blood and bone marrow of the patients administered with the drug-using flow cytometry [103,104]. Another preclinical study evaluated the potential use of follistatin (FST) as a pharmacodynamic biomarker. It was seen that, in FLT3-ITD patients treated with quizartinib, serum FST levels significantly decreased but resurged during relapse [105].Another study identified that the expression of immune checkpoint markers CD155 and CD112 (using flow cytometry and real-time PCR) was specifically downregulated upon treatment with gilteritinib and quizartinib in FLT3-mutated cell models. Thus, CD155 and CD112 have the potential to serve as PD biomarkers for FLT3-ITD AML patients [17].Although FLT3 inhibitors show response in AML patients, the duration of this response is short-lived due to primary and acquired resistance. The most common mechanism of acquired resistance in patients is due to on-target mutations in the tyrosine kinase domain. F691L and D835 are frequently occurring FLT3 gatekeeper mutations. These mutations hinder the drug binding which results in an active kinase conformation unfavorable to interaction with FLT3 inhibitors [100,106]. This resistance mechanism was reported for type II inhibitors, including quizartinib and sorafenib. Both gilteritinib and crenolanib had preclinical and clinical activity against FLT3 D835 mutations, but they had limited activity against the F691L mutations [83,107,108,109]. However, pexidartinib and ponatinib had activity against F691L mutations in preclinical models [98]. Recently, another FLT3 inhibitor, FF-10101, displayed significant activity against F691L and D835 both in vitro and in vivo [110].In addition, FLT3-ITD mutations contribute to resistance to the FLT3 inhibitors. This is because FLT3-WT is sensitive to FLT3 ligand and resistant to FLT3 inhibitors. FLT3-ITD has a WT sequence; therefore, it contributes to the resistance [111,112,113].It was also observed that high levels of FL in the bone marrow microenvironment during induction and consolidation therapy can lead to activation of the FLT3-MAPK pathway and provide a survival signal to the blast cells even in the presence of FLT3 inhibitors [111]. Some preclinical studies also demonstrated that CYP3A4 in the bone marrow stromal cells also leads to FLT3-TKI resistance [114].Acquired resistance due to non-overlapping, secondary mutations is caused by different FLT3 inhibitors. In-vitro-based studies demonstrated that SU5614 produced TK2 changes in D835 exclusively; however, midostaurin produced mutations in TK1 at N676. In addition, sorafenib produced resistant mutations in TK1 (F691L) and TK2 (Y842). These mutations led to different drug responses. While TK2 mutations were sensitive to midostaurin, sunitinib, and sorafenib, TK1 mutations had a differential response to SU5614, sorafenib, and sunitinib but impaired response to midostaurin [115].Additional resistance mechanisms were seen where FLT3-TKI resistant cells became FLT3-independent due to the activation of parallel signaling pathways, including Ras/MEK/MAPK and PI3K/Akt, which compensate for cell survival signals when FLT3 is inhibited. Additionally, activating mutations in the Ras/MAPK pathway, including NRAS, PTPN11, KRAS, and CBL, were of common occurrence in gilteritinib and crenolanib resistance [116,117,118].Given that FLT3 inhibitors present limited efficacy due to the reasons mentioned above, alternate approaches are required to cure FLT3-AML patients. One of the approaches is the use of combination therapies of FLT3 inhibitors with other agents to enhance their efficacy and identify synergistic drug combinations. A combination of sorafenib with vorinostat (histone deacetylase inhibitor) was seen to be effective against FLT3 AML in an early-phase clinical trial [106]. Additionally, triple combinations of sorafenib, vorinostat, and bortezomib (proteasome inhibitor) were effective in early-phase clinical trials [119]. Similarly, a combination of sorafenib and azacytidine (DNA methyltransferase inhibitor) was effective for patients with FLT3-ITD and relapsed AML [120]. Combination therapies with signaling proteins downstream of the FLT3 pathway are another viable approach to overcome FLT3-inhibitor resistance. It was seen that pimozide (STAT5 inhibitor) is synergistic with midostaurin and sunitinib in FLT3-ITD patients in early-phase clinical trials [121]. Recently developed PIM1/FLT3 dual inhibitor SEL23/MEN1703, which targets Pim-1 (a kinase downstream of FLT3) and FLT3 together, is currently undergoing clinical trials [103]. Another study is assessing the safety of everolimus (mTOR inhibitor) with midostaurin in patients with relapsed and refractory AML (NCT00819546). A promising approach is targeting heat shock proteins, including Hsp40, Hsp70, and Hsp90 [122,123,124,125]. Hsp70 inhibition suppresses the proliferation of FLT3-ITD-positive and drug-resistant AML cells via the induction of proteasome-mediated degradation of FLT3-ITD [126]. Co-treatment of midostaurin with 17-AAG (Hsp90 inhibitor) attenuated phospho-FLT3 and induced apoptosis in human acute leukemia cells MV4-11 [127].Some studies used anti-FLT3 monoclonal antibody (mAb) treatment for FLT3-TKI-resistant clones, but it was found to be a cytotoxic [128,129,130]. An alternative approach would be using FLT3 inhibitors with anti-FLT3 antibodies and/or inhibitors of the downstream signaling proteins of FLT3 [128].Additionally, chimeric antigen receptor (CAR) T cells are an emerging novel therapeutic approach to target FLT3 AML. CD8+ and CD4+ T-cells expressing an FLT3-specific chimeric antigen receptor (CAR) were efficacious in vitro. Treatment with crenolanib enhanced the surface expression of FLT3 on FLT3-ITD AML cells which led to increased recognition by FLT3-CAR T cells in preclinical studies [131].FLT3 inhibitors show promising efficacies in progressive and relapsed AML, but the duration of the clinical response is short. Biomarkers are of great importance in predicting the biological behavior of AML, as well as monitoring the efficacy of FLT3 inhibitors in patients. The expression of biomarkers can be used to predict the disease activity in real time. They can also serve as a guide for administering a particular FLT3 inhibitor to achieve CR and high OS.Additionally, to overcome the resistance to FLT3 inhibitors caused by different mechanisms, combination therapies of FLT3 inhibitors with other targeted agents or immunotherapeutic approaches are additional areas of investigation.Conceptualization, N., J.W. and A.-M.H.; software, N.; formal analysis, N., J.W. and A.-M.H.; investigation, N., J.W. and A.-M.H.; resources, N., J.W. and A.-M.H.; data curation, N; writing—original draft preparation, N., J.W. and A.-M.H.; writing—review and editing, N., J.W. and A.-M.H.; visualization, N., J.W. and A.-M.H.; supervision, N., J.W. and A.-M.H. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The authors declare no conflict of interest.Structure of FLT3 and its drug targets. The structure of FLT3 in its inactive conformation which, upon binding to FLT3 ligand (FL), becomes active, resulting in its autophosphorylation. Different FLT3 inhibitors and their binding sites on their domains.Analysis of FLT3 expression in AML patients. (A) Transcript levels of FLT3 in AML patients versus control patients. (B) Percent survival of high-FLT3-expressing patients versus low-FLT3-expressing AML patients. The hazard ratio is 1.8, and the p-value is 0.035, as analyzed from the TCGA dataset upon GEPIA analysis.FLT3 signaling pathway. FL binds to the FLT3 receptor and induces receptor dimerization and conformational changes. FLT3 autophosphorylation activates intracellular signaling cascades including RAS/RAF/MAPK PI3K/AKT/mTOR and JAK/STAT. These pathways control cell proliferation, survival, and apoptosis. These different proteins can be used as predictive, prognostic, and pharmacodynamic biomarkers.Acute myeloid leukemia and acute leukemias of ambiguous lineage (WHO, 2017).FLT3 inhibitors and their targets.Biomarkers for FLT3 AML.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01165.txt

@@ -0,0 +1 @@
+Hepatocellular carcinoma (HCC) is one of the deadliest human health burdens worldwide. However, the molecular mechanism of HCC development is still not fully understood. Sex determining region Y-related high-mobility group box (SOX) transcription factors not only play pivotal roles in cell fate decisions during development but also participate in the initiation and progression of cancer. Given the significance of SOX factors in cancer and their ‘undruggable’ properties, we summarize the role and molecular mechanism of SOX family members in HCC and the regulatory effect of SOX factors in the tumor immune microenvironment (TIME) of various cancers. For the first time, we analyze the association between the levels of SOX factors and that of immune components in HCC, providing clues to the pivotal role of SOX factors in the TIME of HCC. We also discuss the opportunities and challenges of targeting SOX factors for cancer.Sex determining region Y (SRY)-related high-mobility group (HMG) box (SOX) factors belong to an evolutionarily conserved family of transcription factors that play essential roles in cell fate decisions involving numerous developmental processes. In recent years, the significance of SOX factors in the initiation and progression of cancers has been gradually revealed, and they act as potential therapeutic targets for cancer. However, the research involving SOX factors is still preliminary, given that their effects in some leading-edge fields such as tumor immune microenvironment (TIME) remain obscure. More importantly, as a class of ‘undruggable’ molecules, targeting SOX factors still face considerable challenges in achieving clinical translation. Here, we mainly focus on the roles and regulatory mechanisms of SOX family members in hepatocellular carcinoma (HCC), one of the fatal human health burdens worldwide. We then detail the role of SOX members in remodeling TIME and analyze the association between SOX members and immune components in HCC for the first time. In addition, we emphasize several alternative strategies involved in the translational advances of SOX members in cancer. Finally, we discuss the alternative strategies of targeting SOX family for cancer and propose the opportunities and challenges they face based on the current accumulated studies and our understanding.Transcription factors (TFs) are one of the most pivotal sequence-specific DNA-binding proteins responsible for decoding DNA sequences. In order to accomplish the transcriptional regulation of genes, TFs require at least one DNA-binding domain (DBD) for recognizing and binding specific DNA sites located in promoters or enhancers/silencers of targeted genes and another effector domain for recruiting transcriptional machinery to DNA [1]. To this point, over 2000 TFs divided into various families based on the homologous DBDs have been identified, many of which enable the controlling of some critical biological processes in a way that depends on cell type and development patterning [2]. These TFs, which account for about 8% of all human genes, are essential for maintaining normal physiological processes [3]. Consequently, mutation or dysregulation of many master TFs usually leads to a wide range of diseases, and approximately 19% of human TFs are involved in the process [3]. Among them, the tumor is the most concerned disorder.The first oncogenic TFs were found in human acute leukemias [4]. The genes encoding TFs produce fusion proteins with functional alteration due to frequent chromosomal translocations such as PLZF-RARα [5]. These abnormal TFs fusion products disrupt the transcriptional networks during the specific periods of hematopoietic cell differentiation and ultimately cause the initiation of cancer [5]. Additionally, a large number of TFs as drivers or suppressors in solid tumors have been identified. For instance, the MYC TFs are among the most commonly dysregulated driver proteins in an overwhelming majority of human cancers [6]. Upregulated MYC promotes the initiation and progression of cancers by controlling cancer cell-intrinsic oncogenic signaling pathways, tumor microenvironments (TME), and immune surveillance [6]. Similarly, many other TFs such as HIFs, FOXO, and STAT3/5 are also the master regulators that determine the growth, metastasis, and other hallmarks of cancers [7,8,9]. Accordingly, plenty of explorations related to TFs are essential to deeply understand the molecular mechanisms of the initiation and progression of cancers and are beneficial in developing novel cancer therapy strategies. Recently, the sex-determining region Y (SRY)-related high-mobility group (HMG) box (SOX) family TFs, which have been described as playing an essential role in a variety of cancers and as potential therapeutic targets for cancers, is of particular interest.In this review, we first summarize the SOX family in terms of basic structure, classification, and physiological functions. We will focus on the SOX family in hepatocellular carcinoma (HCC), which is one of the deadliest human health burdens worldwide [10], and provide an updated overview of how SOX members are regulated and how they work on the HCC development by controlling the transcription networks. Given that the tumor immune microenvironment (TIME) markedly affects the development and therapy response of tumors, we also detail the role of SOX members in remodeling TIME and analyze the association between SOX members expression and immune components such as immune cells, immune checkpoints, antigen-presenting molecules, chemokines, and chemokine receptors in HCC for the first time. In addition, we emphasize several alternative strategies involved in the translational advances of SOX members in cancer, and finally discuss the opportunities and challenges of targeting the SOX family in cancer based on the current accumulated studies and on our understanding.In 1990, John Gubbay et al. first discovered the SOX family TFs in mammals, given the presence of an amino acid domain characterized by the HMG box of homology with mammalian testis-determining factor SRY [11]. The HMG box, an evolutionarily conserved DNA-binding motif, is the representative signature of the SOX family. This HMG motif comprises 79 amino acids, but only the sequence RPMNAFMVW is conserved for all SOX factors except SRY [12]. This sequence appears to be the most extensive signature that can identify SOX genes [12]. Under long-term selection pressure, this core and ancestral HMG motif gradually changed in sequence but retained the sequence-specific DNA-binding function, coupled with changes in proteins outside the HMG domain, leading to more than 20 distinct SOX proteins in vertebrates [12].On the basis of the degree of sequence identity of the HMG motif, SOX members are classified into eight subgroups, including subgroup A (SRY), subgroup B1 (SOX1, SOX2, and SOX3), subgroup B2 (SOX14 and SOX21), subgroup C (SOX4, SOX11, and SOX12), subgroup D (SOX5, SOX6, SOX13, and SOX23), subgroup E (SOX8, SOX9, and SOX10), subgroup F (SOX7, SOX17, and SOX18), subgroup G (SOX15), and subgroup H (SOX30) (Figure 1). In general, the members within the same SOX subgroup have more than 80% of sequence identity in the HMG motif and are also consistent in other conserved sequences [13]. Accordingly, SOX proteins in the same subgroup equip with similar biochemical properties and probably perform repetitive functions [13]. For instance, all members of the SOX D subgroup, in addition to being highly consistent in the HMG motif, also have the characteristics of an abundant and conserved leucine zipper motif, which contributes to forming homodimers or heterodimers [14]. Consequently, SOX5 and SOX6 proteins of the SOX D subgroup show overlapping and cooperative functions in chondrogenesis [15]. On the other hand, many SOX proteins from different subgroups play distinct and even opposed functions though all SOX proteins, recognizing and binding to a common DNA consensus motif [16]. Past studies have revealed that SOX proteins can recognize many different DNA binding sites. The possible factors that give rise to this difference include flanking sequences that impact binding affinity, SOX proteins dimerization, and interaction between SOX proteins with other TFs due to their adjacent binding sites [13]. Additionally, whether the SOX protein possesses a transactivation domain (such as SRY, SOX B1, SOX C, SOX E, SOX F, and SOX G) or a transcriptional repressor domain (such as SOX B2) determines the transcriptional direction [13]. The post-translational modifications of SOX proteins also affect their transcriptional activity [17]. All these factors lead to the difference and diversity of SOX protein functions.SOX proteins that play essential roles in cell fate decisions involving numerous developmental processes, including neural development, skeletal development, testis development, endoderm development, vascular/lymphatic development, and hair follicle development have been sufficiently studied [18]. Stem cells, as a source of tissue and organ formation that support the growth and development of the body, are usually regulated by context-specific TFs to control their self-renewal and differentiation [19]. Accumulated evidence indicates that the SOX proteins participate in controlling stem and progenitor cell fate and tissue regeneration. By interacting with intracellular cofactors and competing with other TFs of alternative lineages, SOX proteins transcriptionally activate self-renewal-related genes and inhibit differentiation-related genes, therefore reprogramming and maintaining stem cell populations [19]. In addition, SOX proteins also serve as pioneer TFs, firstly combining and preparing target genes for timely activation to maintain cell self-renewal once differentiation occurs [20,21]. Consequently, SOX factors are key regulators of cell fate, and their alteration and dysregulation often lead to various diseases, especially cancer.The initiation and progression of HCC are controlled by a complex and huge molecular regulatory network. TFs in this network integrate and process extracellular and intracellular signaling pathways and output the key commands in the nucleus to affect the malignant phenotype of HCC. Currently, SOX factors, as pivotal regulators in both tumor driver and suppressor signaling pathways, have been reported to control the proliferation, migration, angiogenesis, cancer stem cell (CSCs) properties, and drug resistance of HCC extensively. (Figure 2)The SRY gene, which locates on the Y chromosome, initiates testicular differentiation and induces masculinization of mammalian embryos. The protein expression of SRY is limited to the pre-Sertoli cells of male mice 10.5–12.5 days after intercourse and is regulated by DNA methylation-mediated epigenetic silencing [22]. Of note, there is evidence that the copy number of the SRY gene (Yp11.3) is gained or amplified in 11.8% of male patients with HCC, indicating that SRY may promote HCC development [23]. An early study showed that SRY is highly expressed in HCC cells and tissues and is associated with the poor survival of patients with HCC [24]. However, although SRY is a sex-determining gene, there is no difference in SRY expression between male and female HCC patients. Functionally, SRY promotes the migration and invasion of HCC cells [24]. Simultaneously, S. Murakami et al. reported that SRY promotes the proliferation, invasion, and tumorigenicity of two male mice HCC cell lines by directly activating Sgf29 partly through histone H3 acetylation [25]. Their further study extended to human HCC cell lines and discovered that SRY maintains the CSCs properties, including self-renewal, chemoresistance, and tumorigenicity, by transactivating the classic CSCs marker OCT4 and up-regulating other CSCs markers such as CD13, thereby contributing to HCC progression [26]. In 2017, the role of SRY in hepatocarcinogenesis was verified in vivo for the first time [27]. This research team found that SRY is overexpressed in approximately 84% of male HCC, while most female HCC tissues show as SRY negative [27]. In their in vivo experiment, the aggravating liver cell injury, increased cell proliferation, and accelerated tumor growth were observed in both male and female hepatocyte-specific SRY overexpressing transgenic mice followed by DEN administration [27]. The promotion of SRY in hepatocarcinogenesis is partly attributed to the activation of the SOX9 and PDGFRα/PI3K/Akt pathway, given that these pathways are involved in promoting inflammation, fibrosis, and development of HCC [27,28,29,30].Taken together, the SRY functions as an oncogene to accelerate hepatocarcinogenesis and facilitate HCC development. However, its role and regulatory mechanisms in HCC, especially the difference between male and female HCC, remains to be further explored in light of its particular role in sex-determination for males.The SOX B group is divided into the SOX B1 subgroup and the SOX B2 subgroup. In general, the members of SOX B1 play a transcriptional activation role on their targets, while those in SOX B2 play a transcriptional repression role on theirs [31]. SOX1, SOX2, and SOX3 belong to the SOX B1 subgroup. However, these three members display distinct effects in HCC development. SOX1 as a tumor suppressor gene has been demonstrated in HCC. By hypermethylation of its promoter, SOX1 is frequently downregulated in HCC [32]. SOX1 significantly inhibits the proliferation and invasion of HCC through competing with TCFs to bind β-catenin, subsequently inactivating the Wnt signaling pathway [33]. Moreover, SOX1 induces cellular senescence in Hep3B cells [33]. Conversely, SOX2 and SOX3 play roles in promoting the initiation and progression of HCC. It is well known that a unique subpopulation of cells in HCC is present, whether liver CSCs or tumor-initiating cells (TICs), which share features with normal stem or progenitor cells, including self-renewal and differentiation [34]. These liver CSCs/TICs regenerate all malignant phenotypes of tumors due to their stemness-related capabilities, therefore facilitating the development, recurrence, and therapy resistance of HCC [34]. SOX2, as a transcriptionally regulatory center of self-renewal in embryonic stem cells, plays a core role in reprograming adult somatic cells into a pluripotent stem cell-like state [35,36]. Furthermore, studies showed that SOX2 is highly expressed in HCC, and its overexpression is correlated with the poor survival of patients with HCC [37,38]. Consequently, the cancer-promoting effect performed by SOX2 in HCC is likely to be mainly attributable to its ability to control the stemness of HCC cells. Indeed, well-documented evidence links the role of SOX2 in stemness regulation and HCC promotion. Specifically, a study uncovered that Cyclin G1 up-regulates the expression of SOX2 by activating the Akt/mTOR signaling pathway, which leads to an increased proportion and enhanced tumor-initiating capacity of liver TICs, as well as inducing chemoresistance of HCC cells to sorafenib [39]. Also of interest is that blocking the Akt/mTOR signaling pathway by inhibitors significantly enhances the sensitivity of HCC cells to sorafenib, providing a potential combination therapeutic strategy for HCC [39]. Indeed, drug resistance has severely limited the efficacy of sorafenib. As a tyrosine kinase inhibitor (TKI) that mainly acts by inhibiting the RAS/RAF/MEK/ERK signaling pathway, the main neuronal isoform of RAF, BRAF, and MEK pathways and BRAF mutation are important resistance mechanisms of sorafenib [40]. Therefore, combination therapy strategies to eliminate these key resistance factors are the current priority. The activated Akt/mTOR signaling pathway has been demonstrated to confer sorafenib resistance to HCC through multiple mechanisms, including induction of Warburg shift [41] and increased TIC features [42]. The finding that SOX2-mediated the Akt/mTOR signaling pathway induces stemness and sorafenib resistance in HCC complements the stemness-related molecular mechanisms of sorafenib resistance and further confirms the combination therapy of the Akt/mTOR signaling pathway inhibitor and sorafenib for HCC is feasible. In addition, SOX2 is the critical downstream effector of SIRT1, which enhances the self-renewal ability of CSCs and induces HCC development [43]. Mechanistically, SIRT1 deacetylates the histones and interacts with DNMT3A to demethylate the CpGs in the SOX2 promoter, leading to an upregulation of SOX2 transcription [43]. Moreover, Zhang et al. revealed that TGFβ negatively regulates SOX2 in tumor-initiating hepatocytes [44]. According to their study, SOX2 mediates the suppressive role of TGFβ on HCC development through directly transcriptionally activating lncRNA H19 [44]. As we know, the epithelial-mesenchymal transition (EMT) is a cell-biological program that converts the epithelial characteristics into mesenchymal properties, leading tumor cells to lose adhesion and apical-basal polarity, ultimately promoting the invasion and metastases of the tumor [45]. Studies showed that the EMT process induces CSCs properties in tumor cells, indicating that these two cellular states are closely associated [45]. Consistent with this, SOX2, as a key regulator of liver CSCs, also induces the EMT process in HCC [37,46]. Additionally, SOX2 also involves other mechanisms to contribute to the HCC development, including mediating p38a-ROS-induced suppression of hepatocarcinogenesis and activating the CCAT1/EGFR/miR-222-5p/CYLD signal axis to promote HCC progression [38,47]. SOX3 is reported to be significantly overexpressed in HCC tissues and associated with HCC development, recurrence, and poor prognosis of patients with HCC [48]. However, its role and related mechanisms in HCC remain unclear, although it acts as an oncogene in many other cancers [49,50].Among the members of the SOX C subgroup, SOX4 is the most studied in HCC. Convincing evidence has shown that SOX4, as the direct transcription target of Smad2 and Smad3 downstream of TGFβ, plays a master role in TGFβ-induced EMT, invasion, and metastasis of tumors [51,52]. In HCC, the SOX4 expression is remarkably increased in tumor tissues, especially in metastatic and recurrent HCC samples [53,54,55,56,57,58,59]. The overexpressed SOX4 is positively associated with higher intratumoral microvessel density, distant metastasis, and poor survival of HCC cases, which provides a potential diagnostic and prognostic marker for HCC [56,57,58,59]. As a crucial metastasis-associated regulator, SOX4 induces the chemotaxis of human umbilical vein endothelial cells, angiogenesis, and tumor growth in HCC by activating CXCL12, therefore facilitating HCC metastasis [58]. Treatment with AMD3100, an antagonist targeting CXCR4 (CXCL12 receptor), suppresses the chemotaxis and tube formation of endothelial cells mediated by overexpressed SOX4 [58]. It was also found that SOX4 is overexpressed in liver TICs and plays an indispensable role in liver TICs self-renewal [56]. Mechanistically, SOX4 directly interacts with lncSox4 and STAT3 in the same region of its promotor, thereby up-regulating SOX4 expression and contributing to the liver TICs self-renewal [56]. In addition, as a partner regulator, SOX4 interacts with p53 and specifically impairs the transcription capability of p53 to its downstream target Bax, thus inhibiting the apoptosis of HCC cells following irradiation [54]. Interestingly, although SOX4 acts as a transcription activator in most cancers, its transcriptional repression role has been found in two HCC cell lines [53]. In the future, it is necessary to clarify the distinct function presented by the inverse transcriptional activity of SOX4 in tumors. Past studies have also mined the upstream modifications of SOX4, especially post-transcriptional regulation mediated by microRNA (miRNA). So far, a variety of miRNAs, including miR-129-2, miR-449 family, miR-130a-3p, miR-363-3p, and miR-138 have been suggested to directly or indirectly repress the SOX4 expression in HCC, therefore affecting HCC progression and metastasis [55,57,60,61,62].Hepatitis B virus (HBV) infection remains the primary risk factor for the initiation and development of HCC [63]. There is evidence that HBV replication is closely related to SOX4 levels and forms a positive regulatory feedback loop [64]. First, HBV up-regulates the SOX4 expression at multiple levels: HBV induces the MAPK/YY1 signaling pathway and subsequently directly transactivates SOX4; HBV up-regulates SOX4 at the post-transcriptional level through abolishing the inhibition of miR-335, miR-129-2, and miR-203 on SOX4 expression; at the post-translational level, HBsAg, the surface antigen of HBV, blocks the SOX4 polyubiquitin by interacting with SOX4 to protect it from proteasome-mediated degradation [64]. Meanwhile, viral-induced SOX4 overexpression, in turn, enhances HBV replication through the interaction between viral genomic DNA and the HMG domain of SOX4 in HCC cells [64]. However, Shu Shi et al. found that the binding sequence (AACAAAG) of SOX4 on HBV proposed by the study mentioned above is only present in 9.17% of all HBV genotype strains [65]. In addition, the conclusion that SOX4 facilitates HBV replication contradicts the previous finding that SOX4 is overexpressed in HCC, while HBV replication levels are low in HCC [66,67]. Accordingly, they further explored the regulatory correlation between SOX4 and HBV and came to an opposing viewpoint that SOX4 represses most HBV replication through inhibiting HNF4α expression, which does not involve the binding sequence of SOX4 on the HBV genome [65]. These inconsistent findings indicate that the intricate and precise role of SOX4 on HBV replication remains not to be fully elucidated, and further research is needed which will provide insight into the molecular mechanism of hepatocarcinogenesis from the etiology.Accumulated evidence showed that SOX11 is overexpressed in most cancers, including mantle cell lymphoma, glioma, medulloblastoma, ovarian cancer, and breast cancer [68,69,70,71,72]. The key pro-oncogenic role of SOX11 has also been demonstrated in various cancers [73,74]. However, SOX11 serves as a tumor suppressor to participate in HCC progression. Recent studies suggested that the SOX11 expression level is aberrantly downregulated in HCC tissues relative to paired adjacent noncancerous tissues [75,76]. Mechanistically, SOX11 dampens cell proliferation, induces cell cycle arrest, promotes cell apoptosis, and enhances chemosensitivity in HCC through up-regulating NLK1 expression to inactivate the Wnt/β-catenin signaling pathway [76]. In addition, a study also revealed that SOX11, as the direct downstream target of miR-9-5p, is up-regulated by lncMEG3 and positively mediates the role of LncMEG3 in suppressing HCC growth and promoting HCC cell apoptosis [75]. However, given the little evidence so far, the exact function of SOX11 remains controversial.SOX12 is a novel marker for liver CSCs and is related to tumor metastasis [77,78]. Our previous study unveiled that SOX12 is evidently up-regulated in HCC tissues, especially in HCC metastasis tissues [79]. The overexpressed SOX12 is associated with tumor encapsulation, microvascular invasion, higher TNM stage, and poor prognosis of HCC cases, suggesting that SOX12 could be an independent risk factor for recurrence and worse survival [79]. SOX12 induces the EMT process by directly up-regulating Twist1 expression and facilitates the invasion and metastasis of HCC through transactivating the Twist1 and FGFBP1 expression [79]. Regarding its upstream regulatory mechanism, we disclosed that SOX12 is directly transcriptionally activated by FoxQ1 and plays an indispensable role in FoxQ1-induced HCC metastasis [79]. Additionally, SOX12 is the direct target of miR-874, miR-744, and miR-296-5p to mediate their effects on suppressing the progression and metastasis of HCC [80,81,82].According to current studies, the members of SOX D exhibit different roles in HCC. SOX6 functions as a tumor suppressor and mediates various miRNAs to affect HCC progression. For instance, miR-155, miR-96, miR-19a-3p, and miR-376c-3p directly bind and repress the SOX6 expression to regulate key molecules or pathways such as the Wnt/β-catenin pathway, thereby promoting HCC progression [83,84,85]. It is interesting to note that SOX6 is the target of miR-1269. However, the single-nucleotide polymorphism (SNP) rs73239138 in miR-1269 interferes with the specific binding to 3′ untranslated region (3’UTR) of SOX6, thus preventing miR-1269 from inhibiting SOX6 [86]. Furthermore, the SOX6 level in HCC tissues is relatively low, and the SOX6 expression is negatively associated with the tumor stage and the serum AFP level [87]. The low SOX6 expression predicts shorter disease-free survival and overall survival, indicating that SOX6 is a potential prognostic marker for HCC [87].On the contrary, SOX5 and SOX13 play a cancer-promoting role in HCC. The high-expressed SOX5 and SOX13 are frequently observed in HCC tissues compared with paired nontumor tissues [88,89]. In particular, SOX5 is overexpressed in HCC cases with portal vein tumor thrombosis (PVTT), suggesting that it may be involved in the HCC metastasis [88]. Indeed, a former study demonstrated that SOX5 promotes the migration, invasion, and EMT process of HCC in vitro, which is probably attributable to the up-regulation of Twist1 expression [88]. On the other hand, the overexpressed SOX13 is remarkably associated with poor differentiation, metastasis, recurrence, and worse survival of HCC cases [89]. Generally speaking, polymeric complexes are the frequent form of SOX proteins that play their role on a common target [90,91]. A previous study has shown that dimerization of L-SOX5 and SOX6 are co-expressed with SOX9 and act together on the promoter of the chondrocyte differentiation marker Col2a1 [92]. In HCC, SOX13 enables it to dimerize with its partner SOX5 via the coiled-coiled domain, and functionally cooperate to activate Twist1 transcription, thereby promoting the migration, invasion, and EMT process of HCC [89]. In addition, SOX13 also regulates the stem-like properties of HCC cells [93]. Specifically, SOX13 facilitates the proliferation, self-renewal, tumor-initiating, and chemoresistance of HCC cells through inducing TAZ transcription, given that TAZ is an essential effector of CSC properties [93].SOX9, which is expressed in the progenitor or precursor cells of the embryonic liver and pancreas, serves as a marker of stem cells and progenitor cells in the liver and pancreas [94]. In HCC, SOX9 is also a crucial CSC marker, given that SOX9 endows the classic CSCs characteristics to HCC cells, including tumorsphere formation and resistance to sorafenib [95,96,97]. Simultaneously, the molecular mechanisms underlying SOX9 induces CSC properties have also been discovered. Chungang Liu et al. uncovered that SOX9 enables increasing symmetrical cell division and enhancing the stemness of liver CSCs via suppressing the Numb expression, a key Notch signaling antagonist that promotes Notch degradation [29]. Symmetrical and asymmetrical cell division are the two main types of cell division during proliferation and development to maintain the self-renewal of cells [29]. Asymmetrical cell division generates one stem cell and one differentiating cell to sustain stem cell homeostasis, while symmetrical cell division leads to two daughter stem cells to augment the stem cell pool [98]. Evidence has suggested that tilting the balance toward symmetrical cell division remarkably elevates the number of CSCs in various tumors [99,100,101]. Herein, this finding in HCC not only expands the tumor pool where symmetrical cell division increases CSCs but also reveals a novel mechanism related to SOX9-induced CSCs properties in HCC. Additionally, SOX9 maintains the liver CSCs properties by transcriptionally up-regulating FZD7 to activate the Wnt/β-catenin signaling pathway [95]. The TGF-β/Smad signaling is another important pathway involved in SOX9 regulating the CSCs features [102]. A previous study revealed that SOX9 directly transactivates circular RNA Circ-FOXP1 to sponge miR-875-3p and miR-421 and activate their respective downstream targets, including SOX9, therefore forming a positive feedback oncogenic loop and leading to the growth and metastasis of HCC [103]. As a feature of CSCs, SOX9 enhances sorafenib resistance potentially through promoting CSCs phenotypes under hypoxic conditions or activating the Akt/ABCG2 pathway [96,97]. Of note, hypoxia, as a crucial hallmark of HCC, is responsible for sorafenib resistance to HCC. The mechanisms underlying hypoxia-induced HCC cells escape from sorafenib include HIF-mediated metabolic reprogramming [104], regulation of the PI3K/AKT signaling pathway [105], etc. The above-mentioned study on SOX9 further expands the mechanism of hypoxia-induced sorafenib resistance and closely links hypoxia, CSCs, and sorafenib sensitivity.In the adult liver, the SOX9 expression is mainly observed in the cholangiocytes lining the bile ducts, although low SOX9 expression is also detected in hepatocytes surrounding the ductular structures [94]. Nevertheless, substantial studies demonstrated that SOX9 is overexpressed in HCC tissues and is closely related to poor differentiation, venous invasion, high tumor stage, and worse survival of patients with HCC [29,95,106,107,108]. This phenomenon indicates a presence of aberrant upstream modifications that give rise to the dysregulated SOX9 in HCC. Indeed, various miRNAs, including miR-101, miR-138, miR-1-3p, miR-613, miR-5590-3p, miR-520f-3p, and miR-206, have been verified to target SOX9 directly and inhibit its expression [96,107,109,110,111,112,113]. The PTEN is a classic tumor suppressor, which is lost or decreased in approximately 50% of HCC and is frequently phosphorylated-mediated inactivated in 89% of tumor cases [114,115,116]. Intriguingly, in a model with the Pten deletion in Sox9+ cells within the liver, heterogeneous tumors consisting of HCC cells and cholangiocarcinoma cells or bile duct adenoma cells are developed [117]. These tumor cells originate from Sox9+ cells, suggesting that Pten deletion induces the transformation of the Sox9+ cells [117]. Moreover, following liver injury induced by a high-fat diet or chemical administration, the number of Sox9+ cells expanded significantly, and the formation of these mixed-lineage tumors was remarkably accelerated [117]. Furthermore, the Wnt/β-catenin signaling pathway is needed to maintain the proliferation, survival, and self-renewal of the Pten null SOX9+ cells [117]. Consequently, Pten deletion can induce the transformation of SOX9+ cells into CSCs, and liver injury activates the transformed SOX9+ cells to promote proliferation and ultimately boost hepatocarcinogenesis [117]. Besides, the mRNA and protein stability of SOX9 is also regulated in HCC. CD73, as a novel marker for liver CSCs, in addition to activating the AKT signaling pathway to promote the SOX9 transcription by c-MYC activation, also protects SOX9 from proteasome-mediated degradation by GSK3β inhibition, therefore maintaining the liver CSCs properties [118]. Similarly, lncDUXAP9 improves the mRNA stability of SOX9 and up-regulates its expression by binding to the 3’UTR of SOX9, thereby sustaining the stemness of HCC [119].To our knowledge, over 90% of patients with HCC arise in the setting of chronic liver disease, and cirrhosis or liver fibrosis is the most common risk factor [120]. Hepatic stellate cells (HSCs) are an essential driver of liver fibrosis. Their activation state interplays with various resident cells and tumor cells in the TME via releasing multiple cytokines or profibrotic factors to contribute to the growth and metastasis of HCC [121,122,123]. However, the mechanisms underlying the HSCs that facilitate the growth and invasion of HCC remain elusive. Yu Chen et al. disclosed that the up-regulated SOX9 and INHBB in HCC cells promote the growth and metastasis of HCC through activating HSCs presented in the TME [108]. Mechanistically, SOX9 binds to the enhancer of INHBB and induces its expression, thereby increasing the secretion of activin B from HCC cells into the microenvironment, leading to the activation of peri-tumoral HSCs mediated by activin B/Smad signaling and subsequently promoting liver fibrosis and metastasis of HCC [108]. This finding expands the interaction mechanism of HSCs activation on HCC metastasis and provides a potential therapeutic strategy for targeting the communication between HSCs and tumor cells in the TME.To date, evidence related to the role of SOX8 and SOX10 in HCC is rare. Only one study suggested that the mRNA levels of SOX8 and SOX10 are increased in HCC tissue compared to adjacent benign tissues and healthy tissues [124]. However, the protein level of SOX8 is negatively associated with that of SOX10 in HCC tissues [124]. Functionally, SOX8 promotes the proliferation of HCC cells and activates the Wnt/β-catenin pathway [124]. From this, further related study is needed.In general, SOX7 functions as a tumor suppressor in most tumors. However, its expression level varies greatly among different tumor types, suggesting the existence of cancer cell-dependent mechanisms to regulate SOX7 expression [125]. In HCC, SOX7 is significantly down-regulated relative to adjacent non-tumor tissue and is associated with the advanced stage of HCC [126,127]. The low SOX7 expression predicts the poor prognosis of patients with HCC, which serves as an independent prognostic factor for HCC [127]. As a tumor-suppressor factor, SOX7 inhibits HCC cell proliferation and induces cell cycle arrest by decreasing the expression of cyclin D1 and c-myc [126]. Zhiyun Zheng et al. found that SOX7 is the target of miR-452 [128]. According to their report, miR-452 maintains the stemness of HCC by targeting SOX7 and directly binding to the β-catenin/TCF/LEF transcriptional factor complex to relieve the suppression of SOX7 to the Wnt/β-catenin signaling pathway [128]. Moreover, miR-184 and miR-935 have also been identified to target SOX7 directly and inhibit its expression [129,130]. Nevertheless, the understanding of the upstream regulatory mechanism of SOX7 remains obscure so far.In concordance with the SOX17 in colorectal cancer [131], in HCC, SOX17 is frequently methylated at its promoter region and occurs in approximately 82% of HCC samples [132]. SOX17 inhibits the proliferation of HCC cells and inactivates the WNT/β-catenin signaling pathway through the HMG region [132].The remarkably elevated expression of SOX18 has been observed in HCC tissues compared to that in adjacent nontumor tissues, particularly in cases of metastasis or recurrence [133,134]. Moreover, increased SOX18 expression is correlated with poor tumor differentiation, higher TNM stage, and worse prognosis of patients with HCC, which serves as an independent predictor for survival and recurrence [133,134]. Functionally, SOX18 promotes the proliferation, cell cycle process, EMT process, invasion, and migration and inhibits apoptosis of HCC [133,135]. SOX18 promotes cell viability partly through regulating the AMPK/mTOR signaling pathway [135]. On the basis of these findings, our previous study further found that SOX18 promotes the metastasis of HCC by activating the transcription of FGFR4 and FLT4 directly [134]. The overexpression of SOX18 in HCC can be attributed to the mechanism by which FGFR4 and its ligand, FGF19, activate the p-FRS2/p-GSK3β/β-catenin pathway to transactivate the SOX18 promoter directly [134]. Therefore, a positive feedback loop of FGF19/SOX18/FGFR4 is formed and facilitates the HCC metastasis [134].SOX30 is the only member of the SOX H subgroup. To this point, the tumor suppressor role of SOX30 has been detected in multiple tumors, including bladder cancer and lung cancer [136,137]. In HCC tissues, the expression of SOX30 is decreased compared to adjacent non-tumor tissues [138]. Consistently, SOX30 interferes with cell proliferation and induces cell apoptosis by transactivating its downstream p53 directly in HCC [138]. Furthermore, SOX30 is also a target of miR-645 to mediate the role of miR-645 in promoting HCC progression [138].The TIME is a heterogenous ecosystem consisting of various innate and adaptive immune cells [139]. Substantial evidence suggests that suppressive TIME attributed to the crosstalk between different cells through cytokines, chemokines, and immunosuppressive checkpoints is frequently present in tumors and prominently accounts for the tumor initiation, progression, and therapeutic resistance [139,140,141]. It is becoming increasingly evident that SOX factors play a role in remodeling TIME by impacting multiple immune cells, including effector CD8+ T cells, neutrophils, B cells, and myeloid-derived suppressor cells (MDSCs).Type I interferon (IFNI) signaling, in addition to fighting viral infection, also promotes the recruitment and activation of effector T-cells, activates antigen-presenting cells (APCs), and facilitates the cross-priming of dendritic cells (DCs) in the TME, therefore playing an essential role in natural and therapy-induced cancer immunosurveillance [142,143]. In head and neck squamous cell carcinoma, experimental data suggest that SOX2 induces immunosuppressive TME through disturbing the stimulator of interferon genes (STING)-mediated IFNI signaling activation [144]. In detail, SOX2 negatively regulates the transcription of IFNB1 and IFNI target CXCL10 induced by STING agonist cGAMP and intracellular poly (dA:dT), and accelerates autophagy-mediated STING degradation, leading to a decrease of CD8+T cells infiltration and a relative increase of PD-1high CD8+T cells population, which represents an immune exhaustion state, eventually contributing to tumor growth [144]. Of note, the immune landscape of tumor specimens showed that the SOX2 expression is positively associated with the regulatory T cell (Treg) infiltration and is negatively related to M1-like macrophages, which is consistent with the fact that IFNI promotes M1-like polarization of APCs [144]. Consequently, this SOX2-IFNI axis is a pivotal signal to mediate tumor immune escape. Furthermore, the SOX2/IFN-mediated tumor immune escape has also been demonstrated in melanoma. Ruiyan Wu et al. reported that SOX2 inhibits the transcription of SOCS3 and PTPN1, further activates the JAK-STAT signaling pathway, and up-regulates the interferon-stimulated genes resistance signature (ISG.RS) to hinder the infiltration and cytotoxicity of CD8+T cells [145]. Evidence showed that IFNI signaling also induces resistance to immune checkpoint inhibitors (ICIs) by activating ISG.RS [146,147]. In their study, they found that high SOX2 expression is associated with worse survival and low objective response rate in patients with high PD-L1 who were treated with the anti-PD-1 monoclonal antibody, which indicates that SOX2 is an independent predictor for poor prognosis and resistance to anti-PD-1 treatment in melanoma patients with high PD-L1 levels [145]. By developing genetically engineered mouse models of non-small cell lung cancer, Gurkan Mollaoglu et al. found that the overexpressed SOX2 and the inhibition of NKX2-1 by SOX2 synergistically promote tumor-associated neutrophil (TANs) recruitment through up-regulating CXCL5 expression, facilitating adeno-to-squamous transdifferentiation and squamous tumorigenesis [148]. Additionally, in glioblastoma stem-like cells, SOX2 and OCT4 cooperate to promote immunosuppressive TME by activating immunosuppressive transcriptome mediated by the BRD4/H3k27Ac axis, which includes various immunosuppressive checkpoint molecules (i.e., PD-L1, CD70, A2aR, TDO), cytokines, and chemokines involved in T-cell apoptosis, Treg infiltration, and M2 macrophage polarization [149]. This finding provides a clue to the broader role of SOX2 in constructing immunosuppressive TME. However, conflicting results indicate that SOX2 inhibits the expression of PD-L1 during stem cell differentiation and lung cancer cell plasticity [150]. Consequently, the specific immunosuppressive roles and mechanisms of SOX2 in different tumor types remain to be elucidated.In a triple-negative breast cancer (TNBC) model, SOX4 enables the induction of the resistance of cancer cells to cytotoxic T cells through regulating several innate and adaptive immune pathways, including the suppression of IFNI-stimulated genes and MHC class I pathway genes (HLA-A, HLA-B, and TAP1) [151]. It has been reported that TGFβ induces SOX4 expression [152,153]. In this research, a mechanism study found that ITGAV, which encodes integrin αv protein, up-regulates SOX4 expression by activating TGFβ from a latent precursor [151]. Therefore, the αvβ6–TGFβ–SOX4 pathway is essential in conferring cancer cell resistance to T cell-mediated cytotoxicity and serves as a promising therapeutic target for cancers. In addition, a previous study also revealed that SOX4 is a negative target of miR-132 in B cells, promoting B cell development and increasing the potential of B cell cancer occurrence [154].Through comparatively analyzing the impact of distinct genetic backgrounds in prostate cancer on the composition of immune cells within TME, Marco Bezzi et al. uncovered that the loss of Zbtb7a up-regulates the CXCL5 expression by inducing SOX9 transcriptional activity, thus increasing the recruitment of polymorphonuclear MDSCs in the Pten-deficient tumors and facilitating tumor progression [155]. Likewise, Col1 deletion in aSMA+ myofibroblasts promotes the recruitment of CD206+F4/80+Arg1+ MDSCs through the SOX9-CXCL5 axis and further inhibits T and B lymphocyte function, thereby accelerating pancreatic ductal adenocarcinoma progression [156]. Consequently, the SOX9-CXCL5 axis represents a key regulatory mechanism that mediates the MDSCs chemotaxis to tumors and provides a possible therapeutic strategy for different tumors.So far, the regulatory role of SOX factors in the TIME of HCC is still completely unclear. Herein, we initially explored the association between SOX expression and immune components in HCC using the TIMER2 database. As shown in Figure 3, the expression of most SOX members was significantly associated with the infiltration of eight immune cells in HCC, including CD8+ T cell, CD4+ T cell, B cell, macrophage (M1 and M2 types), MDSC, Treg, neutrophil, and DC. Among them, the tumor-associated macrophage (TAM), MDSC, Treg, regulatory dendritic cell (DCreg), neutrophil, and regulatory B cell (Breg) are essential immunosuppressive cells that determinate the initiation and progression of tumors and as potential therapeutic targets for tumors [157,158,159]. Our analysis results showed that SOX4, SOX11, SOX13, and SOX15 had the most significant and comprehensive positive correlations with these immunosuppressive cells in HCC. Next, we performed gene co-expression analysis between SOX members and immune-related genes in HCC. Via linking the mRNA levels of SOX factors with that of 47 well-known immune checkpoint genes in HCC [160], we found that most SOX members except SRY, SOX1, SOX3, SOX10, and SOX14 were significantly positively correlated with these immune checkpoints (Figure 4A). Of note, some members have shown remarkable correlation with classic immune checkpoints that have successfully implemented clinical translations, including PD-1 (encoded by PDCD1), PD-L1 (encoded by CD274), and CTLA4 (encoded by CTLA4). In addition, we also revealed that the expression of many SOX members was observably associated with that of human leukocyte antigen (HLA)-I and II molecular [161] (Figure 4B), chemokines (Figure 5A), and chemokines receptors (Figure 5B) in HCC.Taken together, the expression correlation between SOX factors and immune components was close in HCC, suggesting that SOX factors were likely to participate in TIME remodeling of HCC and might act as potential biomarkers for predicting immunotherapy response in HCC. Although the crucial role of SOX factors in regulating the initiation and progression of tumors has been well documented, directly targeting these SOX proteins through inhibitors remains a challenge. The explanation for this difficulty is that, unlike the tractable catalytic sites of kinases, TF functions via protein-DNA or protein-protein interactions to form convex and highly positively charged DNA binding interfaces or to flatter protein binding surfaces, losing the structure of the deep pocket in kinases [4]. This structure makes the TFs ‘undruggable’ for traditional small-molecule inhibitors based on structure design. In this section, we summarize several alternative strategies for targeting SOX with translational potential in cancer therapy and detail the experimental evidence in multiple tumors (Table 1).An alternative approach is to use SOX as a biomarker for patient stratification treatment; in other words, to target the upstream or downstream signal molecules of SOX for patients with high SOX levels.In melanoma, the SOX2-BRD4 transcriptional complex has been verified to activate GLI1 via a noncanonical hedgehog/GLI signaling [162]. Smoothened (SMO) is a receptor that mediates downstream GLI1 activation in canonical hedgehog/GLI signaling. However, targeting SMO is frequently ineffective in multiple tumors due to the activation of noncanonical hedgehog/GLI signaling. Recently, researchers combined the SMO inhibitor MRT-92 and a Proteolysis Targeted Chimeras (PROTACs)-derived BRD4 degrader (MZ1), to treat the tumor and found that it remarkably suppressed tumor growth [162]. Simultaneously, high levels of SOX2 serve as a biomarker to achieve precise subpopulation selection for this promising therapeutic strategy [162].In previous research, we have introduced that the integrin αvβ6–TGFβ–SOX4 pathway promotes immune evasion and tumor progression in TNBC [151]. They further found that integrin αvβ6-blocking antibody monotherapy or combined with a PD-1 antibody, which has entered clinical trials, significantly down-regulates the SOX4 expression, sensitize cancer cells to killing by CD8+ T cells, and reduce the primary tumor burden and lung metastatic burden [151]. This finding not only provides a promising immunotherapy strategy but also implies that SOX4 is a biomarker to guide this immunotherapy.SOX9 also acts as an indicator for subpopulation treatment. Evidence showed that the WNT/β-catenin signaling is activated in prostate cancer cells, especially those with SOX9 overexpression, indicating that prostate cancer patients with high SOX9 levels are more sensitive to WNT pathway antagonists [163]. Furthermore, given the molecular mechanism by which SOX9 activates the PI3K/AKT/mTOR signaling pathway in esophageal cancer, the highly expressed SOX9 is a potential indicator for a subset of esophageal cancer patients who are more responsive to rapamycin, a specific mTOR inhibitor that has been used clinically [164].In our previous study, on the basis of the finding that the FGF19-SOX18-FGFR4 positive feedback loop promotes HCC metastasis, we further investigated the effect of BLU9931, a specific FGFR4 inhibitor, in HCC [134]. Results showed that BLU9931 treatment remarkably abrogates the SOX18-induced invasion and metastasis of HCC, suggesting that SOX18 is a biomarker for identifying HCC patients who benefit from FGFR4 inhibitor treatment [134].Though screening the epigenetic compounds library, researchers identified that suberoylanilide hydroxamic acid (SAHA), a histone deacetylase inhibitor that has been approved for the treatment of cutaneous T-cell lymphoma [175], facilitates SOX2 acetylation and proteasome-dependent degradation, therefore abolishing the SOX2-mediated resistance of tumor cells to CD8+T cells [145]. It is worth noting that the administration of SAHA significantly improves the therapeutic efficacy of the anti-PD-1 antibody in a mouse model of melanoma [145]. Currently, several clinical trials evaluating the efficacy of combining SAHA and ICIs are ongoing (NCT02638090, NCT02538510, NCT02619253, NCT02395627). In addition, ChlA-F, a conformation-derivative of Chel A isolated from Goniothalamus cheliensis Hu, promotes SOX2 ubiquitination and protein degradation by enhancing the mRNA stability of E3 ligase USP8 and also inhibits SOX2 protein translation by activating c-Jun-miR-200c, thus suppressing the invasion of bladder cancer cells [165]. However, the efficacy and safety of ChlA-F remain to be investigated in vivo.Choosing appropriate targeted tumor antigens to enhance the anti-tumor immune response mediated by T cells is an effective cancer treatment strategy. Accumulated evidence indicates that several SOX members have been identified as tumor-specific antigens that can augment cytotoxic T lymphocytes (CTLs) response to kill cancer cells, including SOX2 (glioblastoma) [166], SOX4 (lung cancer) [167], SOX6 (glioblastoma) [168,169], and SOX11 (glioblastoma) [170]. In general, these SOX factors are elected due to their significant overexpression in most tumor tissues and their almost negligible levels in nontumorous tissues.Using small interfering RNAs (siRNAs) for gene therapy has the advantages of easy synthesis, high binding specificity, and significant silencing efficacy [176]. However, the obstacles that include poor stability of RNA oligomers, off-target effects, and inability to cross biological barriers have greatly restricted their applications [176]. Delivery of siRNAs through vehicles is a promising strategy to solve the above impediments. Nanoparticles (NPs) as a carrier further provide advantages over other delivery vehicles [176]. Terrick Andey et al. designed a lipoplex nanoparticle to deliver therapeutic siRNA targeting SOX2 (CL-siSOX2) to a mouse xenograft lung cancer model [171]. They found that it significantly suppresses the growth of the tumor with high SOX2 levels and inhibits the expression of markers involved in tumor growth, metastasis, and chemoresistance [171]. Furthermore, in addition to being well-tolerant to CL-siSOX2, the mice also show fewer side effects and decreased tumor size in the combination treatment group of CL-siSOX2 and cisplatin, which provides an effective SOX2-targeted strategy for lung cancer, either monotherapy or in combination with cisplatin [171]. Additionally, another study improved the delivery system of selenium nanoparticles (SeNPs) and developed more stable RGDfC-modified functionalized SeNPs (RGDfC-SeNPs) to selectively deliver siSOX2 to HCC cells through clathrin-mediated endocytosis and releasing siSOX2 in the lysosome [172,177]. This RGDfC-SeNPs/siSOX2 complex exhibits a significant suppressive role on HCC with low toxicity in vitro and in vivo, indicating its potential for HCC-targeted therapy via silencing SOX2 [172].Artificial transcription factors (ATFs) are a molecular tool with great potential that can regulate endogenous target gene expression. Among them, zinc-finger (ZF)-based ATFs enable the specific and effective modulating of target genes expression in vitro and in vivo, which provides additional advantages [178]. In a previous study, researchers engineered three ATFs that bind to the proximal SOX2 promoter and one ATF that targets SOX2 regulatory region I (SRR1), one of which inhibits the expression of SOX2 in breast cancer cells by as much as 94% [173]. Some of these ZF-based ATFs significantly inhibit the growth of breast cancer in vitro and in vivo for a long time [173]. Mechanistically, these ZF-based ATFs play an inhibitory role on SOX2 expression through recruiting the transcriptional repressor domain Kruppel-Associated box (SKD domain) to the SOX2 promoter and further recruiting the co-repressor KAP1 to facilitate chromatin condensation mediated by histones deacetylation and H3K4me3 as well as H3K9me3 demethylation [173,179,180]. Similarly, targeting SOX2 by ZF-based ATFs has also been applied in lung and esophageal squamous cell carcinoma and remarkably inhibits the growth of tumors in vitro and in vivo [174]. Taken together, ZF-based ATFs are an effective molecular-targeted therapy to potentially address the ‘undruggable’ nature of SOX.Over the past decades, not only the role of the SOX family in cell fate decisions has been elucidated, but also its relevance to the initiation and progression of tumors has become gradually clear. Substantial evidence reveals that the SOX family regulates the malignant phenotypes of various tumors extensively, including proliferation, migration, angiogenesis, CSCs properties, EMT, and drug resistance [181]. In HCC, in addition to impacting these phenotypes that are common in other tumors, SOX factors are also involved in the regulation of HCC-specific hallmarks. For example, HBV interacts with SOX4 in HCC, although it is still controversial whether SOX4 regulates HBV replication positively or negatively [64,65]. From a molecular mechanism standpoint, SOX factors play their tumor-promoting or tumor suppressing role in HCC through modulating multiple key signaling pathways in a transcription-dependent or independent manner, including Wnt/β-catenin signaling, TGFβ signaling, Notch signaling, AMPK/mTOR signaling, and p53 signaling. As the primary regulator of body development, the Wnt/β-catenin signaling pathway is the most extensive mediator that affects SOX factors to regulate HCC, especially mediating the role of SOX in regulating the liver CSCs properties [95,128]. Additionally, the aberrant expression of SOX factors is frequently observed in HCC tissues, and some of them act as potential prognostic factors in tumors. According to current studies, promoter hypermethylation, signaling pathway regulation such as TGFβ signaling, AKT signaling, and FGF19/p-FRS2/p-GSK3β/β-catenin signaling, and post-transcriptional modulation by miRNAs, are the main factors responsible for abnormal SOX expression. In summary, this compelling evidence emphasizes the key role of SOX members in the initiation and progression of HCC, but whether they have unknown functions and molecular mechanisms in HCC still needs to be explored.The TIME is characterized by heterogeneity and suppression, and it acts as a key and the hottest factor affecting the initiation and development of tumors. In recent years, the relationship between SOX members and TIME has been gradually revealed. These SOX factors serve as important regulators to mediate the interaction between cancer cells, mesenchymal cells, and immune cells such as effector CD8+ T cells, neutrophils, B cells, and MDSCs, thus remodeling TIME and impacting tumor development. However, the effect of SOX members on TIME in HCC remains unknown so far. In this review, we initially analyzed the association between SOX expression and immune components in HCC by using the TIMER2 database. The exciting results showed that, except for SRY, SOX1, SOX10, and SOX14, the vast majority of SOX members were closely correlated with the infiltration of CD8+ T cell, CD4+ T cell, B cell, macrophage, MDSC, Treg, neutrophil, and DC in HCC (Figure 3). Furthermore, we analyzed the gene co-expression relationship between SOX members and immune checkpoint genes, antigen-presenting molecules, chemokines, and chemokine receptors in HCC and suggested that the expression of most SOX members was implicated in those of immune-related molecules (Figure 4 and Figure 5). In general, our results provided a vital clue that the role of SOX members on HCC extended to affect its TIME and paved the way for exploring the role and mechanism of SOX members in regulating the TIME of HCC by interacting with different immune cells.Many SOX members show significant clinical implications in HCC. For instance, the overexpressed SOX4 or SOX12 is positively associated with the poor survival of HCC, while low SOX6 or SOX7 expression predicts shorter disease-free survival and overall survival, which provide potential diagnostic and prognostic markers for HCC [56,57,58,59,79,87,127]. In addition to serving as the predictors of diagnosis and prognosis, many SOX members are also potential therapeutic targets for HCC. However, despite these wide acknowledgments of the important role of SOX factors on HCC and the encouragement from some successful cases of other TFs in preclinical development or clinical trials (such as STAT TFs), advances in SOX-targeted therapies still face a great challenge. Some alternative strategies with translational potential that include targeting upstream or downstream signal molecules of SOX for patients with high SOX levels, targeting SOX proteins degradation, developing a SOX-related peptide vaccine, delivering siRNA-SOX, and targeting endogenous SOX expression by ATF-based technologies have been proposed in recent years and have obtained promising results in initial preclinical experiments. However, these approaches still have many shortcomings preventing their advancement. Therefore, it is necessary to adopt or develop reasonable and practical approaches for these ‘undruggable’ molecules. For example, the existing methods for SOX proteins degradation lack specificity and stability. A new approach termed PROTACs is designed to degrade polyubiquitin-labeled target proteins through the ubiquitin-proteasome system, thereby avoiding some limitations of traditional small molecule therapy strategies [182]. Although this approach is currently only developed for some ‘druggable’ molecules, it also provides an opportunity for the ‘undruggable’ targets, and future research should make efforts in this regard [183]. Additionally, the tertiary structure of proteins is essential for the development of targeted therapeutic drugs. A recent study found that the intrinsically disordered regions frequently observed in many TFs become structured after being folded and bound to their partners, thereby forming a ‘druggable’ structure [4]. This strategy encourages future research to explore the finer 3D structures of SOX proteins and the interaction between SOX factors and their binding partners.In summary, the significance of SOX TFs in the initiation and development of HCC has been concerned and increasingly uncovered. Given that TIME is critical for tumor initiation and progression, we not only provided a first overview understanding of the remodeling effect of SOX factors on TIME in multiple tumors by summarizing the roles and regulatory mechanisms of SOX factors on immune cells in several tumors in detail, but also revealed the close relationship between SOX factors with eight immune cells and numerous immune-related molecules in HCC by bioinformatic analysis for the first time, which provided an initiatory and latest clue for exploring the effect of SOX factors on TIME in HCC and developing novel combination immunotherapy strategies. As intractable TFs, we concluded a series of alternative strategies being developed in preclinical or clinical trials for targeting SOX in tumors, and proposed some promising alternative strategies, providing some new ideas for SOX-targeted therapy. However, there is still a considerable challenge in turning the targeted SOX factors in cancer from undruggable to reality. Consequently, future work should continue to advance SOX-related research involving molecular mechanisms, anti-tumor immune response, and targeting therapies, such as combination immunotherapy.Conceptualization, L.X.; software, M.S.; validation, X.L. and L.X.; writing—original draft preparation, X.L.; writing—review and editing, X.L. and L.X.; visualization, X.J., M.X. and T.Z.; supervision, Y.W. and W.H.; funding acquisition, L.X. and W.H. All authors have read and agreed to the published version of the manuscript.This research was funded by the National Natural Science Foundation of China No. 81972237 (L.X.), No. 81871911 (W.H.), No. 82173313 (W.H.), and No. 81772623 (L.X.).Publicly available datasets were analyzed in this study. This data can be found here: [timer.cistrome.org/ (accessed on: 25 November 2021) and clinicaltrials.gov/ (accessed on: 28 November 2021)].The authors declare no conflict of interest.Schematic diagram of the chromosomal positions and domain structures of human SOX transcriptional factors. SOX family members are classified into eight subgroups based on the degree of sequence identity of the HMG motif, which is an evolutionarily conserved DNA-binding motif and is also a representative signature of the SOX family. Besides the shared HMG motif, different SOX members have their specific structure domains to achieve functional diversity. The chromosomal positions and domain structures of SOX members have been shown (Data referenced from Ref. [17] and website www.ncbi.nlm.nih.gov/gene/ (accessed on: 15 November 2021)). Abbreviations: SRY: Sex-determining region Y; SOX: Sex-determining region Y-related high-mobility group box; HMG: High-mobility group.The roles and molecular mechanisms of SOX transcription factors in hepatocellular carcinoma (HCC). SOX members serve as master tumor drivers or suppressors to transcriptionally regulate key downstream targets or signaling pathways, therefore controlling the initiation and progression of HCC. Simultaneously, the dysregulated SOX expression is frequently observed in HCC, which is mainly attributed to promoter hypermethylation, signaling pathway regulation, and post-transcriptional modulation by miRNAs (microRNAs). This diagram was created by BioRender.com (accessed on: 28 November 2021). The red box represents tumor driver, and the blue box represents tumor suppressor. Abbreviations: SOX: Sex-determining region Y-related high-mobility group box; SRY: Sex-determining region Y; CSC: Cancer stem cell; 3′UTR: 3′ untranslated region; H3: Histone 3; EMT: Epithelial-mesenchymal transition.The association heatmaps between SOX transcription factor expression and immune cell infiltration in hepatocellular carcinoma. Red indicates positive correlation, and blue indicates negative correlation. The darker the color, the stronger the correlation. The data was analyzed by the TIMER2. * p < 0.05, ** p < 0.01, *** p < 0.001.Correlation between the expression of SOX transcription factors and that of immune-related genes in hepatocellular carcinoma. (A) The relationships between SOX factors expression and the levels of immune checkpoints genes and (B) antigen-presenting molecules were analyzed by the TIMER2. Red indicates positive correlation and blue indicates negative correlation. The darker the color, the stronger the correlation. * p < 0.05, ** p < 0.01, *** p < 0.001.Correlation between the expression of SOX transcription factors and that of immune cell chemotaxis-related genes in hepatocellular carcinoma. (A) The relationships between SOX factors expression and the levels of chemokines and (B) chemokine receptors were analyzed by the TIMER2. Red indicates positive correlation and blue indicates negative correlation. The darker the color, the stronger the correlation. * p < 0.05, ** p < 0.01, *** p < 0.001.The alternative strategies for targeting SOX factors in cancers.Abbreviations: SMO: Smoothened; SOX: Sex-determining region Y-related high-mobility group box; mAb: Monoclonal antibody; TNBC: Triple-negative breast cancer; HCC: Hepatocellular carcinoma; SAHA: Suberoylanilide hydroxamic acid; CTLs: Cytotoxic T lymphocytes; CL-siSOX2: siSOX2 delivered by cationic lipoplex; RGDfC-SeNP: RGDfC-modified functionalized selenium nanoparticles; ZF-based ATF: Zinc-finger-based Artificial transcription factors; SKD: Kruppel Associated box; SCC: Squamous cell carcinoma;.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01166.txt

@@ -0,0 +1 @@
+In the lack of direct comparative evidence of stereotactic body radiation therapy, we reviewed one of the largest locally advanced pancreatic cancer cohort homogeneously treated in a tertiary cancer center. Our propensity score–matched analysis shows comparable outcomes between stereotactic body radiation therapy and concurrent chemoradiotherapy in terms of survival, local control, and treatment-related toxicities. Considering the advantages of SBRT such as short treatment duration, better tolerance, easy combination with systemic treatment, and the potential for dose escalation, further investigation of the feasibility of SBRT as an alternative to CCRT in treating locally advanced pancreatic cancer is required.In locally advanced pancreatic cancer (LAPC), stereotactic body radiation therapy (SBRT) has been applied as an alternative to concurrent chemoradiotherapy (CCRT); however, direct comparative evidence between these two modalities is scarce. The aim of this study was to compare the clinical outcomes of SBRT with CCRT for LAPC. We retrospectively reviewed the medical records of patients with LAPC who received SBRT (n = 95) or CCRT (n = 66) with a concurrent 5-FU-based regimen between January 2008 and July 2016. The clinical outcomes of freedom from local progression (FFLP), progression-free survival (PFS), overall survival (OS), and toxicities were analyzed before and after propensity score (PS) matching. After a median follow-up duration of 15.5 months (range, 2.3–64.5), the median OS, PFS, and FFLP of the unmatched patients were 17.3 months, 11 months, and 19.6 months, respectively. After PS matching, there were no significant differences between the SBRT and CCRT groups in terms of the 1-year rates of OS (66.7% vs. 80%, p = 0.455), PFS (40.0% vs. 54.2%, p = 0.123), and FFLP (77.2% and 87.1%, p = 0.691). Our results suggest SBRT could be a feasible alternative to CCRT in treating patients with LAPC.Pancreatic cancer is an aggressive malignancy, and only 10–20% of newly diagnosed patients are suitable for complete resection, which is considered the only curative approach [1]. Unresectable, or locally advanced pancreatic cancer (LAPC), accounts for about 30% of all pancreatic cancer, and although various combinations of chemotherapy and radiotherapy (RT) have been tried to improve the oncologic outcome, patients with LAPC still have a dismal prognosis with a median survival of 5–15 months [2,3]. The comparison between conventional concurrent chemoradiotherapy (CCRT) and chemotherapy alone showed contradictory results in several large LAPC trials, including the recent LAP07 study [4,5,6,7]. This phase III randomized study failed to show a significant gain in survival after CCRT despite better local control and similar toxicity rates [8]. The authors commented that further intensification of treatment strategies to treat early micrometastatic spread and enable downstaging might be necessary to achieve better oncologic outcomes.Stereotactic body radiotherapy (SBRT) is a modern RT technique that has several advantages over conventional RT and has been widely applied as an effective local therapy for various types of cancers. SBRT enables the conformal and accurate delivery of high radiation doses in a few fractions while minimizing the irradiation of surrounding normal tissues. SBRT is considered to have different tumoricidal mechanisms, such as vascular endothelial destruction and immune modulation [9,10].In LAPC, several studies reported favorable oncologic results of SBRT as an alternative treatment to CCRT [11,12,13,14,15]. However, there are few studies that compared SBRT with conventional CCRT or chemotherapy, and there is still a lack of empirical evidence for this new technique [16,17,18,19]. Therefore, in the present study, we compared the oncologic outcomes of patients with LAPC treated with SBRT or CCRT in order to provide detailed information about the optimal treatment strategy for LAPC.We retrospectively reviewed the medical records of patients with histologically confirmed pancreatic cancer who were treated at our institution between January 2008 and July 2016. The eligibility criteria were as follows: (1) unresectable tumor classified by the multidisciplinary oncology team review of imaging studies according to the National Comprehensive Cancer Network (NCCN) guideline [20]; (2) underwent SBRT or CCRT; (3) no distant metastases at baseline or before RT; (4) Eastern Cooperative Oncology Group (ECOG) performance status of 0 to 2. The initial patient evaluation included physical examination, complete blood count, standard blood chemistry panel including carbohydrate antigen 19-9 (CA19-9), pancreatic protocol computed tomography (CT) scan, chest radiograph, magnetic resonance imaging (MRI), and positron emission tomography-CT (PET-CT) scan. This study was approved by the institutional review board of Asan Medical Center, and informed consent was waived due to the retrospective nature of the study.The use of induction chemotherapy and the method of RT were determined at the discretion of treating physicians. Induction chemotherapy was defined as the start of chemotherapy more than 1 month prior to RT. For SBRT, the respiratory-gated intensity-modulated radiation therapy (IMRT) or volumetric modulated arc radiotherapy (VMAT) technique was used. The SBRT procedure used at our institution was described in our previous papers [21,22]. Briefly, a four-dimensional CT (GE LightSpeed RT 16; GE Healthcare, Waukesha, WI, USA) simulation was performed during free breathing. A Real-time Position Management Respiratory Gating system (Varian Medical Systems, Palo Alto, CA, USA) was used to record the patients’ breathing patterns. The CT data were sorted according to the respiratory phase, and treatment planning was performed based on the CT images at the end-expiratory phase.Both the primary tumor and metastatic regional lymph nodes were included in the gross tumor volume (GTV) when target coverage and dose constraints could be maintained. A lymph node was regarded as metastasis if it was more than 1 cm in short-axis diameter or if it had necrotic features. Diagnostic CT, MRI, and PET-CT images were used to assist in defining the GTV. To reduce internal motion margins, a respiratory gating scheme around the end-expiratory phase (30 to 70% in most cases) was applied to all patients. The maximum intensity projection (MIP) images corresponding to the gating window were consulted to contour the internal target volume (ITV). Three or four gold seeds implanted near the primary tumor or pancreatic duct stent were used as an internal marker, and full-phase trajectory was delineated. The planning target volume (PTV) was defined using 3 mm isotropic margins to the ITV in order to account for set-up errors, unless the margin resulted in expansion into the duodenum or stomach; in such cases, a non-uniform PTV margin expansion was used provided that the GTV dose constraints were met. The prescribed dose was administered to the isodose line covering the PTV.The total dose was mainly determined based on general dosing guidelines after determining the dose to be administered to the normal organs, including the following: maximal point dose to the stomach, duodenum, or small bowel was kept to <30 Gy, and ≥700 cm3 of the normal liver was kept to <15 Gy. The volume of 75% of combined kidneys was kept to <12 Gy, and the maximal point dose to the spinal cord was <20 Gy.All patients in the CCRT group were treated with three-dimensional conformal RT (3DCRT). In addition to the primary tumor and metastatic lymph nodes, the inclusion of regional lymphatics in the clinical target volume was decided by the physician based on the patient’s performance and disease status. A PTV margin of 0.7–1.0 cm was added for daily set-up variations. During RT, all patients received concurrent oral capecitabine or intravenous 5-fluorouracil (5-FU) bolus with leucovorin.After treatment, regular follow-up examinations were performed at 2 to 3-month intervals. Follow-up evaluation included physical examination, complete blood count, standard blood chemistry panel including CA19-9, and an abdominal CT scan. Additional imaging studies were conducted whenever clinically indicated. A contrast-enhanced abdominal CT scan was used for the assessment of treatment response. For patients who responded sufficiently, decision to proceed with surgical resection was made by the multidisciplinary team. Local failure was defined as growth of the radiated pancreatic lesion or regional lymph nodes, and distant failure was defined as clinical or pathological detection of disease beyond the pancreas and regional lymph nodes. During and after treatment, treatment-related toxicities were reported using the Common Terminology Criteria for Adverse Events, version 4.0 (https://ctep.cancer.gov/protocoldevelopment/electronic_applications/ctc.htm, accessed on 23 February 2022). Events reported within 90 days after RT were classified as acute toxicities, whereas those occurring after 90 days were considered late toxicities.Patient characteristics between the two treatment groups were compared by Student’s t-test for continuous variables and the χ2 test or Fisher’s exact test for categorical variables. The Kaplan–Meier method was used to estimate the rates of freedom from local progression (FFLP), progression-free survival (PFS), and overall survival (OS). FFLP was calculated from the date of diagnosis to the date of local failure by radiologic or pathologic examination. PFS was calculated from the date of diagnosis to the date of any progression or death from any cause. OS was calculated from the date of diagnosis to the date of death from any cause. Comparison of survival rates was performed with the log-rank test. Cox proportional hazards model was used to assess the level of statistical significance of prognostic factors for OS. Multivariable analysis was performed with backward elimination of all variables with a p value of <0.2 in the univariate analysis. The cumulative incidence of local recurrence (LR) was estimated and compared using Gary’s test considering death as a competing risk. To identify the risk factors, we used Fine and Gray’s method for modeling the hazard of the sub-distribution to account for death as a competing risk [23].Propensity scores were generated using the logistic regression model that included age, sex, performance status, LN metastasis, tumor abutment to stomach and duodenum, location, size, and pre-RT CA19-9. Greedy matching was performed using a caliper of 0.2 standard deviations of the logit of the propensity score. The absolute standardized differences were used to diagnose the balance after propensity score matching, and all absolute standardized differences were less than 0.15 after matching. All analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA) and SPSS 22.0 (IBM, Armonk, NY, USA).Patient characteristics at the time of radiotherapy are summarized in Table 1. Of the 161 patients, 95 and 66 patients underwent SBRT and CCRT, respectively; 93 (57.8%) patients had a pancreatic head tumor and 114 (70.8%) had clinical N0 disease. Compared with the CCRT group, the SBRT group had a significantly older age (median, 64.0 vs. 60.5, p = 0.045) and higher proportions of patients with tumors larger than 40 mm (66.3% vs. 24.2%, p < 0.001) and CA 19-9 levels higher than 37 U/mL (79.9% vs. 63.6%, p = 0.032). Induction chemotherapy was administered to 40 (42.1%) patients in the SBRT group and 60 (90.9%) patients in the CCRT group (p < 0.001); of the patients who received induction chemotherapy, 32 (32%) received FOLFIRINOX (5-FU, leucovorin, oxaliplatin, and irinotecan) or gemcitabine/nanoparticle albumin-bound (nab)-paclitaxel, while others received other gemcitabine-based regimens. In the SBRT group, 92 (94.7%) patients received 4-fraction (Fx) treatment and the median dose was 28 Gy (range, 24–36). The median dose in the CCRT group was 54 Gy (range, 40–59.4) in 1.8- or 2.0-Gy per Fx. Except for three patients (one with 40 Gy and two with 46 Gy), all patients were treated with doses higher than 50 Gy. Additional chemotherapy was administered to 82 (86.3%) patients in the SBRT group and 16 (24.2%) patients in the CCRT group (p < 0.001). Post-RT chemotherapy regimens were FOLFIRINOX or gemcitabine/nab-paclitaxel in seven patients and other gemcitabine-based regimens in 87 patients.The median follow-up duration was 15.5 months (range, 2.3–64.5), and the median OS, PFS, and FFLP of the total patients were 17.3 months, 11 months, and 19.6 months, respectively. In the SBRT and CCRT groups, the 1-year rates of OS, PFS, and FFLP were 68.4% and 81.8% (p = 0.053, Figure 1A), 42.9% and 53.6% (p = 0.051, Figure 1B), and 80.4% and 80.0% (p = 0.401, Figure 1C), respectively. Multivariate analysis showed that tumor size > 40 mm (hazard ratio (HR), 1.463; 95% confidence interval (CI), 1.034–2.070), curative resection (HR, 0.324; 95% CI, 0.149–0.704), and induction chemotherapy duration > 6 months (HR, 0.572; 95% CI, 0.341–0.958) were significantly associated with OS (Table 2). There were no significant predictive factors for LR in the multivariate analysis (Table 3), and the type of RT was not significantly associated with the OS or LR.By using the propensity score, 45 patients in each group were matched. The baseline characteristics were well balanced between the matched groups, except for the sequence of chemotherapy (Table 4). There were no significant differences between the two groups in the 1-year rates of OS (66.7% vs. 80.0%, p = 0.455, Figure 2A), PFS (40.0% vs. 54.2%, p = 0.123, Figure 2B), and FFLP (77.2% vs. 87.1%, p = 0.691, Figure 2C). The cumulative incidence of LR at 1 year was not significantly different after considering death as a competing risk (22.4% vs. 12.9%, p = 0.906).During a median duration of 3.2 months (range, 2.7–5.6) after the end of RT, seven (7%) patients in the SBRT group and two (3%) patients in the CCRT group underwent curative resection (p = 0.311; Table 5). Images of a representative patient who underwent curative resection after SBRT are presented in Figure 3. The median OS of these nine patients was 28 months (range, 16–65), and R0 resection was achieved in six patients. After surgery, three patients showed distant recurrence, and another three patients showed simultaneous local and distant recurrence. Two (one in each group) patients were still alive at the last follow-up without recurrence. During the follow-up period, 79 (83.2%) patients in the SBRT group and 56 (84.8%) patients in the CCRT group experienced disease progression or recurrence, and the first sites of failure were predominantly distant in both groups (Table 5). Twelve patients (12.6%) in the SBRT group and 17 patients (25.8%) in the CCRT group had an isolated local recurrence (p = 0.102).There were no significant differences between the two groups in the occurrence of acute and late toxicities (Table 5). For acute toxicities, one patient in the SBRT group had nausea and abdominal pain 2 months after RT, and abdominal CT showed peritoneal-free air, suggesting gastric ulcer perforation; the patient was managed with conservative care, and peritoneal air was not evident in the follow-up CT. One patient in the CCRT group stopped the treatment at 50.4 Gy due to duodenal ulcer bleeding at the tumor infiltration site. Bleeding was controlled with conservative treatment, but the patient died of peritoneal tumor progression after 2 months. Another patient in the CCRT group presented with abdominal pain 2 months after treatment, and the abdominal CT showed sealed duodenal perforation. Because the perforation occurred near the duodenal stent that had been inserted before treatment, it was not considered to be related to treatment and was managed conservatively. Other cases of acute toxicities such as nausea and anemia all improved soon after the treatment.For late toxicities, duodenal ulcer bleeding occurred in two patients in the SBRT group and one patient in the CCRT group at 4–6 months after RT; because, the tumors were attached to or invading the ulcer site, it was difficult to determine the causal relationship between treatment and ulcer. Likely owing to their tumor-related nature, these lesions waxed and waned during follow-up. No patients died due to treatment-related toxicity.RT has been widely used for treating LAPCs and has been shown to be effective in achieving local control while preventing pain and obstructive symptoms that deteriorate the quality of life. However, regarding survival benefit, the role of conventional RT has shown contradictory results, and SBRT has been investigated as a potential alternative. In the present study, we reviewed one of the largest LAPC cohorts homogeneously treated in a tertiary cancer center and performed a comparison between SBRT and CCRT. The median survival duration of 17.3 months and the 1-year local control rate of 70–80% in our study were comparable to the previous results of phase III trials and meta-analyses [4,6,24]. After PS matching, there were no significant differences in the rates of survival, local control, and treatment-related toxicities between SBRT and CCRT.Theoretically, SBRT could provide a greater local tumor control and lower toxicities through precise treatment delivery; accordingly, several studies have reported favorable oncologic outcomes of SBRT in LAPC [12,13,14,25,26]. However, high-quality evidence comparing CCRT and SBRT is still lacking. Park et al. compared unmatched 44 SBRT and 226 CCRT patients and reported similar disease control rates, in which the 1-year rates of OS and local failure were 56.2% vs. 59.6% (p = 0.75) and 34.4% vs. 30.2% (p = 0.51) in the two groups, respectively [17]. Lin et al. reported a small study comparing 20 SBRT and 21 IMRT patients, in which there was no significant difference in the 1-year OS (80.0% vs. 70.7%, p = 0.127), while SBRT was associated with better local control in multivariate analysis [16]. A recent Italian multicenter case-control study compared 40 matched pairs of SBRT and CCRT patients and showed that the SBRT group had a non-inferior OS (1 year, 79.8% vs. 73.8%, p = 0.470) but superior local control (1 year, 80.4% vs. 53.1%, p = 0.017) to the CCRT group [27]. In addition, a few meta-analyses and registry studies using the National Cancer Data Base reported that SBRT was associated with improved overall survival compared with CCRT [18,19,28].Regarding treatment-related toxicities, the treatments were well-tolerated in both groups, and there were no significant differences in the incidence of severe (≥grade 3) acute and late toxicities. An Italian study similarly reported that there were no significant differences in the acute and late gastrointestinal toxicities between the SBRT and CCRT groups [27]. However, in both studies, the rate of acute toxicity was higher in the CCRT group, albeit without statistical significance. On the other hand, Park et al. [17] showed a significantly higher rate of acute grade 2 gastrointestinal toxicity in the CCRT group compared with the SBRT group (24% vs. 7%, p = 0.008). Similarly, in a meta-analysis of 9 SBRT and 11 CCRT investigations, severe acute toxicities were more prevalent in the CCRT group than in the SBRT group (37.7% vs. 5.6%, p = 0.0002), while there was no significant difference in severe late toxicities (10.1% vs. 5.9%, p = 0.49).In the absence of a randomized trial, it is difficult to conduct a fair comparison between SBRT and CCRT. Many confounders, such as disease extent, tumor location, induction chemotherapy, socioeconomic status, and type of treatment center, could influence the choice of treatment method. Our propensity score-matched study provides the same result as previous reports that SBRT leads to non-inferior outcomes compared with CCRT. However, unlike several other previous studies, we did not observe significantly superior results in the SBRT group in terms of local control, survival, or toxicity. The reason for the lack of a significant difference might be that our cohort size was not sufficiently large or that confounding variables such as tumor location and tumor-bowel abutment were more well-controlled than in other studies. Further investigation is required to determine whether SBRT could provide improved local control, survival, or toxicity compared with CCRT.SBRT is considered to have additional advantages in the current treatment strategy. The failure of several previous randomized trials in demonstrating the survival benefits of RT was partially due to the fact that distant metastasis was the dominant primary pattern of failure in LAPC patients [4,6,18]. In the present study, the first site of recurrence was a distant location in more than 60% of patients. In such patients, improvements in local control are unlikely to significantly improve their survival. However, recently applied intensified chemotherapy regimens such as FOLFIRINOX and gemcitabine/nab-paclitaxel have demonstrated significantly improved response rates and longer survival in pancreatic cancer [29,30,31,32,33,34]. As systemic control improves, local control can play a more critical role in the patients’ oncologic outcomes; therefore, the roles of RT and RT methods need to be re-evaluated. In this regard, SBRT would be more advantageous than CCRT as it minimizes chemotherapy interruption with short treatment duration and better patient tolerance. Most patients in our SBRT group started post-SBRT chemotherapy within 2 weeks, and previous studies also reported the feasibility of chemotherapy resuming 1 week after SBRT [16,25,26,35].As several studies have demonstrated the dose-response relationship for RT in pancreatic cancer [24,36], a higher dose of RT is desirable for better tumor control. However, administering high-dose RT to the pancreas is difficult as it is surrounded by radiosensitive organs such as the stomach and duodenum. Several different dose and fractionation schemes have been attempted, and some of them that used single Fx were associated with intolerable severe GI toxicities [11,14]. Accordingly, multi-fraction treatment has been widely used in recent and ongoing trials as well as ours [37]. These multi-fraction regimens are currently considered feasible in terms of efficacy and toxicity; however, further dose escalation is suggested when using SBRT with recently developed RT techniques such as MR-guided adaptive RT or simultaneous integrated boost [38,39,40]. Another potential benefit of dose escalation is converting an unresectable LAPC to a resectable one. In the present study, approximately 7% of patients received curative resection after SBRT, which is in line with several previous reports [8]. However, surgical conversion may be a critical issue in future investigations considering that a much higher conversion rate is expected with an intensified chemotherapy regimen [33,41]. To improve the resectability by reducing the tumors around crucial vessels, several ongoing trials prescribe higher doses whenever possible to tumor-vessel interfaces with advanced RT techniques [39,40]. The results of these new RT techniques and higher doses should be followed up.Our study has several limitations. First, the retrospective nature of this study confers potential selection biases. At our institution, there were no policies to skew certain patients towards treatment; however, the selection between SBRT and CCRT was not randomized. Although we conducted propensity score matching to reduce selection biases, several factors, such as number and schedule of chemotherapy, were not included in the propensity score generation and biases could not be completely eliminated. Second, our cohort included treatment-naïve patients as well as those who had received different numbers of induction chemotherapy. Given that we evaluated variables at the time of RT initiation, induction chemotherapy could have affected baseline characteristics and survival estimations. However, in clinical practice, patients who show a good response after induction chemotherapy receive RT, and they usually have a better prognosis than treatment-naïve patients. In the present study, this bias might have worked in favor of the CCRT group, which had a higher proportion of patients who had completed induction chemotherapy than the SBRT group. Therefore, we believe that this bias actually supports the non-inferiority of SBRT. Third, because the majority of the patients were treated before the introduction of FOLFIRINOX or gemcitabine/nab-paclitaxel, these patients did not receive the best treatment options by the current standard. Although our study focuses on RT rather than chemotherapy, the comparison of RT methods with intensified regimens should be investigated in the future. Despite these limitations, to our knowledge, the present study is one of the largest matched studies reporting the non-inferiority of the current fractionated SBRT scheme to CCRT. Further investigations are warranted to confirm our results.In the current study, SBRT was not inferior to CCRT for patients with LAPC in terms of local control, survival, and toxicity. Considering the advantages of SBRT, such as short treatment duration, better tolerance, easy combination with systemic treatment, and the potential for dose escalation, further investigation of the feasibility of SBRT as an alternative to CCRT is required.Conceptualization, J.-h.P., D.-W.S., S.S.L., C.Y.; Methodology, J.-h.P., Y.S.S., H.H.P., S.M.Y., J.J., S.K.; Investigation, J.-h.P., Y.S.S., H.H.P.; Writing—original draft, J.-h.P., Y.S.S., H.H.P.; Writing—review & editing, S.M.Y., J.J., J.H.K., D.-W.S., S.S.L., M.-H.K., S.K.L., D.H.P., T.J.S., D.O., C.Y., B.-Y.R., H.-M.C., K.-p.K., J.H.J.; All authors have read and agreed to the published version of the manuscript.This research received no external funding.The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of Asan Medical Center (2018-1046; approval received on 16 August 2016).Patient consent was waived due to the retrospective, anonymized nature of the study.The datasets used and/or analyzed during this study are available from the corresponding author upon reasonable request.We thank Joon Seo Lim from the Scientific Publications Team at Asan Medical Center for his assistance in preparing this manuscript.The authors declare no conflict of interest.Recurrence and survival outcomes in the unmatched cohort. (A) overall survival (OS); (B) progression-free survival (PFS); (C) freedom from local progression (FFLP).Recurrence and survival outcomes in the propensity score-matched cohort. (A) overall survival (OS), (B) progression-free survival (PFS), (C) freedom from local progression (FFLP).A representative patient who underwent curative resection after stereotactic body radiation therapy (SBRT). (A,B) Pre-SBRT magnetic resonance imaging and computed tomography (CT) show pancreatic head cancer (red arrow) abutting more than 180° of the superior mesenteric artery (SMA). (C) Gross tumor volume and planning target volume in simulation CT. (D) Three months follow-up CT after SBRT shows decreased tumor and SMA abutment.Characteristics of the unmatched cohort at the time of radiotherapy.Abbreviations: SBRT, stereotactic body radiotherapy; CCRT, concurrent chemoradiotherapy; ASD, absolute standardized difference; ECOG PS, Eastern Cooperative Oncology Group performance status; RT, radiotherapy; CA19-9, carbohydrate antigen 19-9; EQD2, equivalent dose in 2 Gy per fraction. Values are presented as median (range) or number (%) of patients.Variables associated with overall survival.Abbreviations: SBRT, stereotactic body radiotherapy; CCRT, concurrent chemoradiotherapy; ECOG PS, Eastern Cooperative Oncology Group performance status; RT, radiotherapy; CA19-9, carbohydrate antigen 19-9; CTx, chemotherapy; EQD2, equivalent dose in 2 Gy per fraction.Variables associated with local recurrence.Abbreviations: SBRT, stereotactic body radiotherapy; CCRT, concurrent chemoradiotherapy; ECOG PS, Eastern Cooperative Oncology Group performance status; RT, radiotherapy; CA19-9, carbohydrate antigen 19-9; CTx, chemotherapy; EQD2, equivalent dose in 2 Gy per fraction.Characteristics of the propensity score-matched cohort at the time of radiotherapy.Abbreviations: SBRT, stereotactic body radiotherapy; CCRT, concurrent chemoradiotherapy; ASD, absolute standardized difference; ECOG PS, Eastern Cooperative Oncology Group performance status; RT, radiotherapy; CA19-9, carbohydrate antigen 19-9; EQD2, equivalent dose in 2 Gy per fraction. Values are presented as median (range) or number (%) of patients.Events after treatment.Abbreviations: SBRT, stereotactic body radiotherapy; CCRT, concurrent chemoradiotherapy. * Toxicity was assessed using Common Terminology Criteria for Adverse Events, version 4.0 (https://ctep.cancer.gov/protocoldevelopment/electronic_applications/ctc.htm, accessed on 23 February 2022). Events reported within 90 days after RT were classified as acute toxicities, whereas those occurring after 90 days were considered as late toxicities.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01167.txt

@@ -0,0 +1 @@
+Ultra-high dose rate radiation, widely nicknamed FLASH-RT, kills tumors without significantly damaging nearby normal tissues. This selective sparing of normal tissue by FLASH-RT tissue is called the FLASH effect. This review explores some of the proposed mechanisms of the FLASH effect and the current data that might support its use in pancreatic cancer. Since radiation for pancreatic cancer treatment is limited by GI toxicity issues and is a disease with one of the lowest five-year survival rates, FLASH-RT could have a large impact in the treatment of this disease with further study.Recent preclinical evidence has shown that ionizing radiation given at an ultra-high dose rate (UHDR), also known as FLASH radiation therapy (FLASH-RT), can selectively reduce radiation injury to normal tissue while remaining isoeffective to conventional radiation therapy (CONV-RT) with respect to tumor killing. Unresectable pancreatic cancer is challenging to control without ablative doses of radiation, but this is difficult to achieve without significant gastrointestinal toxicity. In this review article, we explore the propsed mechanisms of FLASH-RT and its tissue-sparing effect, as well as its relevance and suitability for the treatment of pancreatic cancer. We also briefly discuss the challenges with regard to dosimetry, dose rate, and fractionation for using FLASH-RT to treat this disease.External beam radiation therapy uses ionizing radiation to induce double strand DNA breaks in cancer cells and is thought to selectively kill tumors by exploiting differences in DNA repair between tumors and normal tissues. When radiation therapy is administered to patients, image guidance is used to guide the precise targeting of tumors but cannot eliminate incidental radiation injury to normal tissues. Thus, the doses of radiation used in modern oncology have generally been derived empirically, usually as the maximum permitted by normal tissue toxicity [1]. For some forms of cancer— e.g., early-stage cancers of the lung, prostate, anus, head and neck, and cervix—radiation given with radiosensitizing chemotherapy is sufficient for a cure [2,3,4,5,6]. However, many tumors in the abdomen cannot be sufficiently treated with only radiation and chemotherapy, since the maximum dose of radiation that the intestines can receive over a lifetime is about 50 Gy [7]. Unfortunately, many tumors of the abdomen require much more than 50 Gy to be completely sterilized. Therefore, radiation can be a useful preparative treatment before surgery, but it is often insufficient to achieve oncologic control on its own. Surgery is limited as a treatment option and poses risks such as pain and infection. Chemotherapy shares similar risks of pain as surgery and improvements are limited. Without advancements in radiation protection or radiosensitization, the curative potential of radiation therapy alone for gastrointestinal tumors will be limited.FLASH-RT represents an entirely new paradigm of potentially curative therapy that is broadly applicable across all cancer types. The ability to treat tumors while sparing normal tissues is essentially the ‘holy grail’ of cancer therapy since all oncologic therapies are limited by patient toxicity (chemotherapy and radiation side effects or morbidity from surgery).FLASH-RT encompasses two components: (1) The delivery of ultra-high dose rates (UHDR) and (2) the radiation sparing effect (a.k.a. the FLASH effect). Regarding the former, FLASH-RT is a catchy nickname for the novel radiation treatment that uses ultra-high dose rates to deliver radiation therapy (RT). In conventional radiation therapy (CONV-RT) used in most of the world, the dose rate is generally <0.1 Gy/s, whereas the FLASH-RT dose rate is generally defined as a mean dose rate of ≥40 Gy/s [8,9]. A more precise definition is still being debated and will be important to define in order to ensure scientific rigor and to safeguard the clinical translation of FLASH-RT. Moreover, a more mature definition of FLASH-RT should, at a minimum, include other interdependent parameters, such as dose, dose per pulse, instantaneous dose rate, and total duration of exposure [9,10]. Beyond the ultra high dose rates, the ability for FLASH-RT to selectively spare normal tissues while maintaining tumor killing is what is generating excitement for this new technique. This novel biological phenomenon of selective normal tissue sparing is known as the FLASH effect [11]. The FLASH effect has now been shown repeatedly in many different organ systems, with a dose modifying factor of 1.1–1.8, depending on model species and end point used [11]. However, the mechanism of the FLASH effect has yet to be fully elucidated but is a critical component in the understanding and optimization of FLASH-RT [12]. It is critical to note that the FLASH effect (normal tissue sparing with isoeffective killing of tumors) is what defines FLASH-RT, and thus future trials testing this modality will focus on reducing toxicity. Differences in the extent and type of DNA damage resulting from FLASH- versus CONV-RT could contribute to the observed FLASH effect. The highly energetic ionizing radiation used in-RT is what kill tumors. Specifically, the energy of this type of radiation induces double-stranded DNA breaks that overwhelm the cell’s repair mechanism, leading to cell death. FLASH- and CONV-RT seem to cause different types of DNA damage in normal tissues and tumors [13]. A FLASH-RT dose of 20 Gy has been shown to cause less double-stranded DNA damage to normal cells than CONV radiation does at the same dose [14]. In addition, although the time required for DNA damage repair is similar for CONV radiation and FLASH-RT, UHDR has been shown to produce fewer dicentric chromosomes [14]. Whether the damage that FLASH-RT inflicts on tumor cells differs significantly from CONV-RT remains unknown. Nevertheless, the lesser DNA damage to normal tissue from FLASH-RT than from CONV-RT may partially explain the FLASH effect. Oxygen tension is a critical regulator of radiation sensitivity in classical radiobiological models. Higher oxygen levels have correlated with increased DNA damage, sometimes explained by the ‘oxygen fixation hypothesis’, in which damage induced by radicals is ‘fixed’ by molecular oxygen, and thus cannot be chemically restored. Oxygen depletion has known radioprotective effects on both normal and tumor tissues [15].Consequently, one of the most popular proposed mechanisms of the normal tissue sparing induced by FLASH-RT is known as the oxygen depletion or transient hypoxia hypothesis. Ionizing radiation hydrolyzes cytoplasmic water to yield free radicals that can ultimately cause permanent DNA damage [15]. The processes of downstream formation of radical species and damage fixation are processes that consume oxygen. Thus, a lack of oxygen limits the extent of radiation-induced damage. According to the oxygen depletion hypothesis, the UHDR irradiation creates a temporarily hypoxic environment that confers transient radioresistance in the irradiated tissue (Figure 1) [16,17]. Furthermore, the time interval of delivery (<100–200 ms) is short enough to negate the effect of reoxygenation during the radiation delivery [18]. The phenomenon of transient hypoxia would not occur in CONV-RT because the significantly lower dose rates involved would allow reoxygenation to occur during delivery of the radiation.To achieve the necessary reduction to hypoxic levels without excessive doses of radiation (assuming an oxygen depletion rate of around 0.5 mmHg/Gy) [19], the cells would need to be in physiological, or even close to hypoxic levels, of oxygenation prior to irradiation. Although this not the case for most tissues, some niches of stem cells reside in hypoxic microenvironments, with partial oxygen concentrations as low as 10 mmHg in many different tissue types [20]. The differential effect of normal vs. tumor tissue (normal tissue sparing but iso-effective response on tumor tissue) would then be explained by the already highly hypoxic tumor and would therefore not be affected. Abolfath et al. demonstrates in simulations the significance of different oxygen levels between tumors and normal tissues on the induction of the FLASH effect [21].However, the oxygen depletion hypothesis has been put into question, for several reasons. First, most solid tumors are known to not be uniformly hypoxic, having both hypoxic and non-hypoxic compartments throughout the tumor [22]. Under the transient hypoxia hypothesis, this heterogeneity should then introduce a sparing effect also on tumors, which is not supported by the literature [9]. Second, in vitro studies of normal, non-immortalized cells have shown a sparing effect when irradiated in ambient air conditions (pO2 ≈ 159 mmHg) [23,24,25,26]. Third, simulation studies and oxygen consumption studies in pure water and cell buffer solutions have also challenged this hypothesis by showing that total oxygen depletion did not occur with the use of FLASH radiation [27,28,29]. However, whether total depletion of oxygen is required for the induction of the FLASH effect has yet to be confirmed.Independent of the magnitude of the effect that radiological reduction in oxygen concentration has on tissues after UHDR radiation, oxygen most likely still underlies the FLASH effect. Modulation of the partial oxygen pressure in in vitro experiments had an important effect on the resulting survival fraction and in the increased sparing effect seen after FLASH relative to CONV irradiation [30,31,32]. Similar results have been seen in vivo when increasing the oxygen tension of tissues in mice by having the mice breathe carbogen before and during irradiation, which led to no differences in effects after FLASH versus CONV irradiation [33]. Mice breathing medical air, by contrast, showed strong and reproducible FLASH effects.The role of oxygen is therefore potentially better explained by radical–radical interactions, where high local concentrations of radicals being formed during UHDR irradiation cause the radicals to interact between themselves and effectively ‘quench’ the downstream effects of the produced radicals. Current models on the lifetime of radical formation support this hypothesis. In their models, Labarbe et al. showed the importance of oxygenation level and dose rate in the formation and lifetime of organic peroxyl radicals [34]. Interestingly, moderate oxygenation levels resulted in reduced lifetime of organic peroxyl radicals following UHDR irradiation, while no observed differences were found at low or high oxygenation levels. Other simulation studies on the FLASH effect and its relationship to intracellular oxygen are related to the redox biology specific to normal and tumor tissue [35]. Tumor tissue has an increased ‘normal’ level of reactive oxygen species (ROS) and also has a decreased capacity to handle a rapid increase in ROS levels. This near-saturated system of ROS handling, together with higher levels of labile iron and transferrin receptors in tumor compared to normal tissue, could result in higher induced oxidative damage. Some of these hypotheses are directly testable in an experimental setting, which would be needed to fully elucidate the role of oxygen in relation to the FLASH effect.Oxygen depletion hypothesis of the FLASH effect. Tumors are extremely hypoxic relative to their surrounding normal tissue, with oxygen tensions ranging from 0.3% to 4.2%. The surrounding normal tissue is more highly oxygenated (i.e., physioxia), usually between 3% and 7.4% [36]. For this reason, a more pronounced decrease in oxygen level after FLASH treatment is observed in physioxic conditions compared with tumors. This correlates with a significant decrease in radiosensitivity (or increase in radiation resistance) in normal tissues, conferring the FLASH effect. Tumors, on the other hand, show little change in radiosensitivity, leading to the absence of the FLASH effect.Lymphocytes are among the most radiosensitive tissues and are quickly depleted with prolonged radiation exposure [37]. However, the degree of radiosensitivity of lymphocytes differs depending on whether they are within tumors or are circulating. A recent mouse model study demonstrated that lymphocytes residing in tumors are more radioresistant than are circulating lymphocytes [38]. Specifically, as much as 85% of these resident lymphocytes were preserved compared to 10% of circulating lymphocytes after exposure to 8 Gy whole body irradiation. The same study showed that TGFβ was a key contributor to the increased radioresistance exhibited by tumor-infiltrating lymphocytes. Levels of circulating tumor cells are therefore good markers of the extent of immunocompromise resulting from radiation.This brings us to another popular hypothesis involving modulation of the immune system, known to be crucial in tumor biology and treatment responses. Isolated lymphopenia is practically a pathognomonic feature of conventional irradiation of large areas of the body [38]. FLASH-RT is thought to reduce lymphopenia by decreasing the volume of blood irradiated in any body part (Figure 2). As a hypothetical example, exposure of a large mediastinal radiation field to 4 Gy would entail irradiation of not only many lymph nodes, but also the heart and pulmonary vessels, which oxygenate and recirculate the entire blood volume at a rate of 5 L/min [39]. This modest 4-Gy dose could be delivered with CONV-RT at a typical dose rate of 0.08 Gy/s in 50 s, but even such a modest dose would not only irradiate the mediastinal lymph nodes but also approximately 4L of blood circulating throughout the irradiated volume. In this simple example, CONV-RT would expose nearly the entire circulating blood volume, and radiosensitive lymphocytes in particular, to some dose of radiation. On the other hand, FLASH-RT at dose rates of 40 Gy/s and above would involve exposing only a fraction of the blood volume to radiation, thereby effectively limiting the total exposure of lymphocytes to potentially lethal radiation injury.This hypothetical scenario was supported by a recent mouse simulation study in which exposure to CONV-RT killed 90% to 100% of circulating immune cells as compared with FLASH-RT, which killed only 5% to 10% of such cells in the same treatment setup [40]. That same study showed that immune cell sparing in mice was optimal at FLASH-RT dose rates of at least 40 Gy/s. In contrast, an experimental study showed that radiation delivered at 30 Gy/s failed to protect mice from cardiac and splenic radiation-induced lymphopenia [41]. However, the dose rate of 30 Gy/s may not have been sufficient to induce the FLASH effect in this study. Nevertheless, even at this dose rate, the recruitment of T lymphocytes into the tumor microenvironment was more prevalent after FLASH-RT than after CONV-RT [41]. More studies are needed to better elucidate the contribution of the immune response on the mechanism underlying the FLASH effect.There are several studies showing the promising tissue sparing effects of FLASH-RT. Most have focused on rodent experiments where it has been shown that FLASH-RT induces tissue sparing effects in, e.g., GI tract (crypt regeneration, survival) [42], brain (cognition, neuroinflammation) [43,44], and skin (early skin reactions, necrosis, survival) [45,46,47], compared to CONV-RT. Utilizing rodent models, it has also been shown that the higher therapeutic index of FLASH compared to CONV-RT is due to the reduction in normal tissue toxicity, as the tumor response has been shown to be independent of mode of delivery [48,49]. For instance, one preclinical orthotopic model demonstrated the effectiveness of FLASH-RT in preserving intestinal stem cells to the same degree as in healthy mice, while showing an isoeffective treatment on the ID8 ovarian tumor model compared to CONV irradiation [50].In large animal studies, FLASH-RT was seen to result in lower occurrences of fibro-necrosis in pig skin compared to CONV-RT [51]. In cats with locally advanced squamous cell carcinoma, little to no acute toxicity was found with FLASH-RT [51]. However, it should be noted that other studies using radiation therapy roughly equivalent to FLASH-RT showed no difference in tumor control and no significant tissue sparing [52].To date, only one patient has been treated with FLASH-RT from which the results have been made available to the scientific community. In this case, FLASH-RT was shown to be safe. This patient, a 75 year old male with cutaneous lymphoma, was the first to be treated with FLASH-RT [53]. He had previously underwent localized CONV-RT to treat painful cutaneous lesions but exhibited poor tolerance. FLASH-RT was suggested as a possible radiation treatment that would provide similar tumor control with less toxicity. Treatment with FLASH-RT involved irradiating a 3.5-cm diameter skin tumor on the patient with a 5.6-MeV LINAC capable of delivering FLASH-RT dose rates. The results of the treatment were favorable where the patient exhibited no decrease in skin thickness and an equivalent tumor control compared to previously administered CONV-RT treatments [53]. In a different study on the same patient, two distinct tumors were treated to a dose of 15 Gy single fraction using either CONV or FLASH dose rates [54]. Both treatments gave comparable results in terms of acute and late effects as well as tumor control.Human clinical trials that aim to access the feasibility of FLASH-RT are already underway at the time of this review (February 2022). For instance, Cincinnati Children’s/University of Cinncinati Health Proton Therapy Center initiated a prospective trial under the name FAST-01. The trial began in November 2020 and aims to enroll 10 patients with bone metastases. The main objective of this trial is to determine the feasibility of 1 fraction of 8 Gy FLASH-RT using protons for human radiation treatments [55]. Centre Hospitalier Universitaire Vaudois (CHUV), Lausanne University Hospital, which performed the first human treatment using FLASH-RT, began enrollment for a new FLASH-RT clinical trial in June 2021. Their phase I trial, set to enroll 7 to 21 patients, aims to determine an optimal FLASH-RT dose for improved tumor control in melanoma skin metastases while preventing radiation-induced toxicity [56].Thus, overall, this new technology has been used in humans in a limited capacity, but appears to be safe awaiting further clinical trial data. Based on this promise, we posit potential addition of usage in the setting of pancreatic cancer.The use of FLASH-RT to administer ablative doses of-RT, and therefore the potential for curative therapy, addresses a critical unmet need for patients with unresectable pancreatic cancer. The potential for killing tumors while minimizing toxicity would be perfectly applied to pancreatic cancer, where growth of the primary tumor causes pain and suffering for nearly all afflicted individuals. If FLASH-RT meets even a fraction of its potential in this disease, that would improve the quality of life for patients undergoing treatment and for survivors. With a survival rate of less than 10% at 5 years, even a modest improvement in disease outcomes could double or triple survival rates [57]. Surgery remains the only curative form of treatment, but surgery is an option for only about 10–15% of patients, because most pancreatic cancers present with invasion into nearby blood vessels. Radiation can sometimes be used when surgery is not possible, but its efficacy is limited by the anatomic location of the pancreas (Figure 3a). Moreover, pancreatic cancer is highly resistant to radiation therapy, and therefore ablative doses exceeding 70 Gy are needed for effective tumor control [58]. Unfortunately, the pancreas abuts the duodenum, stomach, and small bowel, which can only tolerate about 50 Gy of radiation. As a result, the dose of radiation that are given to patients with pancreatic cancer is safe but not sufficient to permanently control the disease.Why is targeting the pancreas so difficult? Radiation therapy targets a focused three-dimensional area that must always include a rim of normal tissue (roughly 2–3 mm) to account for changes in patient position and movement of internal organs. Although the pancreas is generally not considered to be mobile within the body, its position can change by as much as 2 cm just from normal respiration. Changes in bowel gas and peristalsis can also alter the amount of normal tissue that happens to be in a radiation field on a given day. Thus, eliminating all normal tissue from a radiation field is not realistic.FLASH-RT is an ideal solution to targeting pancreatic cancer since it can also target the unique biology of pancreatic cancer. A common characteristic of most tumors is a low level of oxygen known as hypoxia. Pancreatic ductal adenocarcinoma (PDAC) is the most common form of pancreatic cancer, marked by a highly hypoxic tumor environment [59], which makes it resistant to both chemotherapy and radiation [60]. Notably, PDAC tumors are extremely hypoxic compared to their surrounding healthy tissue under physioxia (Figure 3b). FLASH-RT may perfectly enable effective hypoxic tumor killing while sparing the healthy intestinal tissue nearby (Figure 3c). This is because the highly hypoxic tumors of PDAC will allow for a more pronounced transient oxygen depletion, enhancing the FLASH effect. Moreover, FLASH-RT would also reduce lymphopenia, which would enable further systemic treatments such as chemotherapy and immunotherapy [61].FLASH-RT would be a completely new weapon against pancreatic cancer because it allows treatments that were previously not possible. For instance, although most ablative techniques target the tumor and a small rim of normal tissue, FLASH-RT could enable elective irradiation of at-risk lymph nodes and at-risk organs that may harbor microscopic disease, such as the liver. FLASH-RT could also be combined with other therapies to further improve its therapeutic ratio. Our group exploited this concept by using selective hypoxia mimicry in normal tissue [62], or tumor-specific radiosensitization [63], to improve the therapeutic ratio of ablative radiation in pancreatic cancer. Finally, existing clinical LINACs could be retrofitted to allow FLASH-RT capability [64], which would enable this type of treatment to be given in many parts of the world that lack new, state-of-the-art radiation delivery machines, which may enhance the possibility of curative treatment.Most FLASH-RT experiments are performed with specialized machines or clinical LINACs that have been converted to produce ionizing radiation at ultra-high dose rates. The generally accepted threshold dose rate of 40 Gy/s is thought to be needed to induce the FLASH effect [65]. Even if a machine could produce this dose rate, there is the issue with properly calibrating the accuracy of the dose delivered. There have been several solutions to this issue of dosimetry of FLASH, which will likely be an important component going forward, particularly for deep tumors seen in pancreatic cancer.Currently, IntraOp Medical has developed an electron-based FLASH LINAC for intraoperative radiation. It has recently been approved for FLASH-RT preclinical experiments and clinical human trials [66]. The use of intraoperative radiation (IORT) in pancreatic cancer has shown prior benefit in local control after resection since it helps to reduce microscopic disease near the retroperitoneal border, which is often challenging to achieve clear margins [67]. However, resectable pancreatic cancer may not be the best first indication for FLASH-RT in pancreatic cancer, since most concerning normal tissue can be manually retracted by surgery during IORT, therefore reducing the need for the sparing effect of FLASH-RT. An electron FLASH approach might be useful in the unresectable setting, particularly if a laparotomy is performed first to expose the intact tumor.Another limitation of studies of FLASH-RT to date has been their use solely as single fractions. On the one hand, fractionation may not be needed in many cases given the apparent normal-tissue-sparing effects of FLASH-RT. On the other hand, however, fractionation may be needed to deliver a fully ablative dose to radiation-resistant tumors such as pancreatic cancer. Stereotactic body radiation therapy (SBRT), in which radiation is given in 3–5 fractions, has been shown to improve outcomes while reducing toxicity through its more accurate targeting [68,69,70]. Whether the FLASH effect is still present if the radiation is given in a fractionated fashion is currently unknown. Therefore, FLASH dose-finding experiments such as these must be conducted in preclinical settings to determine their safety, efficacy, and feasibility before moving on to clinical trials.FLASH-RT is a relatively novel form of radiation therapy that involves the use of ionizing radiation at UHDRs, which distinguishes FLASH-RT from CONV-RT, which involves much lower dose rates. Preclinical studies have shown that FLASH-RT can confer healthy tissue sparing through a phenomenon known as the FLASH effect. The mechanism of the FLASH effect has yet to be fully elucidated, but several hypotheses have been proposed as outlined in this review. Although some contradictions to these hypotheses have been noted, some combination of them—along with other unknown factors—probably participate in creating the FLASH effect. Furthermore, in order to safely bring FLASH-RT to the clinic, the precise definition of what constitutes FLASH-RT is urgently needed.The FLASH effect may be ideally suited for the treatment of pancreatic cancer for several reasons. Pancreatic cancer is characterized by a highly hypoxic tumor environment. Surgery remains the only curative option for localized pancreatic cancer, with radiation often used as an adjuvant after surgery. However, the pancreas’ proximity to the gut and critical blood vessels precludes use of ablative radiation doses in most cases. However, the FLASH effect could help to minimize damage to these normal tissues, which would enable high radiation doses to be administered. FLASH-RT has shown promise in several mouse models and in a few anecdotal clinical situations, but more rigorous study is needed to determine its role in the treatment of pancreatic cancer. The recent expansion of FLASH-RT using intraoperative RT (IORT) equipment opens up the potential for future human trials targeting pancreatic cancer with IORT techniques. This would greatly expand the opportunities for testing FLASH-RT and for further exploring its potential. Considering the relatively low 5-year survival rates for patients with pancreatic cancer, advances in FLASH-RT may prove to be pivotal in significantly improving treatment outcomes in this difficult-to-treat form of cancer.C.M.O., C.M.T. and E.S. were involved in the conceptualization, writing, editing, and figure visualization for this review. Both C.M.T. and E.S. were jointly responsible for project oversight, administration, and funding acquisition. All authors have read and agreed to the published version of the manuscript.C.M.O. is supported by a diversity supplement grant 3R01 CA227517-03S1 from the parent National Institutes of Health (NIH) grant R01CA227517-01A1 and support from the Partnership for Careers in Cancer Science and Medicine from the Office of Faculty Diversity, Equity and Inclusion at the University of Texas MD Anderson. C.M.T. was supported by funding from the National Institutes of Health (NIH) under award number R01CA227517-01A1, the Reaumond Family Foundation, the Mark Foundation, Childress Family Foundation, and generous philanthropic contributions to The University of Texas MD Anderson Moon Shots Program. E.S. was supported by the University Cancer Foundation via the Institutional Research Grant program at MD Anderson Cancer Center, by a grant from MD Anderson’s Division of Radiation Oncology, and in part by Cancer Center Support Grant P30 CA016672 from the National Cancer Institute of the National Institutes of Health, to The University of Texas MD Anderson Cancer Center.We thank Christine F. Wogan, MS, ELS, of MD Anderson’s Division of Radiation Oncology, for careful editing and proofreading assistance in preparing this manuscript. Additionally, the figures were created with BioRender.com (accessed on 18 January 2022).C.M.T. is on the medical advisory board of Accuray and is a paid consultant for Xerient Pharma and Phebra Pty, Ltd. E.S. and C.M.O. have no potential conflicts of interest.Immune hypothesis for the FLASH effect. Lymphocytes circulating within blood vessels are an important component of the immune response that influences tumor suppression. According to the immune hypothesis, the higher dose rates characteristic of FLASH-RT allow exposure of a much smaller volume of blood to radiation than CONV-RT. As a result, a higher number of circulating lymphocytes will survive and the immune response critical for tumor suppression is preserved to a greater degree. In contrast, the lower dose rates used in CONV-RT allow exposure of larger blood volumes circulating through the radiation field, leading to a significant loss of lymphocytes and a compromised immune response.Gastrointestinal organs are susceptible to radiation damage from pancreatic cancer radiotherapy. (a) The duodenal portion of the small intestine, the stomach, and other nearby organs are prone to radiation-induced toxicity, especially the rapidly dividing intestinal cells; (b) Oxygen concentrations (blue wedge) vary between normal tissues and tumors. Pancreatic tumors near the duodenum are especially hypoxic relative to the physioxic environment of the healthy tissue at risk; (c) Nearby normal tissues that would be damaged by CONV-RT would be spared by FLASH-RT, and FLASH-RT would also kill the tumor through the FLASH effect.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01168.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.Ramucirumab is indicated at a dosage of 8 mg/kg every 2 weeks as monotherapy or in combination with paclitaxel for second-line advanced/metastatic gastric/gastroesophageal junction (GEJ) adenocarcinoma. A post hoc analysis of the phase III trials REGARD and RAINBOW suggested a positive correlation between ramucirumab exposure and efficacy. Studies JVDB and JVCZ explored different ramucirumab dosing regimens as monotherapy and in combination with paclitaxel, respectively. Here we report results from these studies, in which JVDB evaluated the pharmacokinetics and safety of the currently registered dosing regimen for ramucirumab monotherapy and three exploratory dosing regimens, and JVCZ evaluated the efficacy and safety of a higher dosing regimen of ramucirumab in combination with paclitaxel in second-line gastric/GEJ adenocarcinoma. Overall, the safety profiles were similar between the registered dose and the exploratory dosing regimens. However, a lack of a dose/exposure-response relationship supports the standard dose of ramucirumab as second-line treatment for patients with advanced/metastatic gastric/GEJ adenocarcinoma.Studies JVDB and JVCZ examined alternative ramucirumab dosing regimens as monotherapy or combined with paclitaxel, respectively, in patients with advanced/metastatic gastric/gastroesophageal junction (GEJ) adenocarcinoma. For JVDB, randomized patients (N = 164) received ramucirumab monotherapy at four doses: 8 mg/kg every 2 weeks (Q2W) (registered dose), 12 mg/kg Q2W, 6 mg/kg weekly (QW), or 8 mg/kg on days 1 and 8 (D1D8) every 3 weeks (Q3W). The primary objectives were the safety and pharmacokinetics of ramucirumab monotherapy. For JVCZ, randomized patients (N = 245) received paclitaxel (80 mg/m2-D1D8D15) plus ramucirumab (8 mg/kg- or 12 mg/kg-Q2W). The primary objective was progression-free survival (PFS) of 12 mg/kg-Q2W arm versus placebo from RAINBOW using meta-analysis. Relative to the registered dose, exploratory dosing regimens (EDRs) led to higher ramucirumab serum concentrations in both studies. EDR safety profiles were consistent with previous studies. In JVDB, serious adverse events occurred more frequently in the 8 mg/kg-D1D8-Q3W arm versus the registered dose; 6 mg/kg-QW EDR had a higher incidence of bleeding/hemorrhage. In JVCZ, PFS was improved with the 12 mg/kg plus paclitaxel combination versus placebo in RAINBOW; however, no significant PFS improvement was observed between the 12 mg/kg and 8 mg/kg arms. The lack of a dose/exposure-response relationship in these studies supports the standard dose of ramucirumab 8 mg/kg-Q2W as monotherapy or in combination with paclitaxel as second-line treatment for advanced/metastatic gastric/GEJ adenocarcinoma.Ramucirumab is a human IgG1 monoclonal antibody that binds to the extracellular domain of the vascular endothelial growth factor receptor-2 (VEGFR-2), blocking binding of VEGF-A, VEGF-C, and VEGF-D [1] and inhibiting tumor angiogenesis [2]. Ramucirumab has shown antitumor activity in phase III trials and has received approval for the treatment of several tumor types, including gastric or gastroesophageal junction (GEJ) adenocarcinomas [3,4,5,6,7]. REGARD [4] and RAINBOW [5] were the pivotal randomized phase III trials that established the safety and efficacy of ramucirumab 8 mg/kg every 2 weeks (Q2W) alone [4] or in combination with paclitaxel [5] in patients with previously treated gastric/GEJ adenocarcinoma.Subsequent exploratory exposure-response analyses of REGARD and RAINBOW data, and a case-control analysis of the latter, suggested higher ramucirumab exposure was associated with longer overall survival (OS) and progression-free survival (PFS) [8,9]. In the exposure-response analysis of RAINBOW, an increased risk of grade ≥3 hypertension, leukopenia, and neutropenia, but not febrile neutropenia, was correlated with increased ramucirumab exposure [8]; however, these toxicities were manageable in the original phase III trials [4,5]. These exposure-response findings, and the fact that the ramucirumab maximum tolerated dose (MTD) is 13 mg/kg/week [2], suggested there may be an opportunity to further improve efficacy while maintaining an acceptable safety profile. Thus, 2 phase II post-marketing commitment studies were conducted: JVDB and JVCZ. Here, we describe the results from these trials, where JVDB evaluated the pharmacokinetics (PK) and safety of the currently registered dosing regimen for ramucirumab monotherapy and three exploratory dosing regimens (EDRs), and JVCZ evaluated the efficacy and safety of a higher dosing regimen of ramucirumab in combination with paclitaxel in second-line gastric/GEJ adenocarcinoma.Studies JVDB (NCT02443883) and JVCZ (NCT02514551) were multinational, open label, randomized, phase II clinical trials that compared alternative ramucirumab doses as monotherapy (JVDB) or combined with paclitaxel (JVCZ) in patients with metastatic or locally advanced gastric/GEJ adenocarcinoma. Eligible patients were aged ≥18 with measurable disease according to Response Evaluation Criteria in Solid Tumors version 1.1 (RECISTv1.1) [10], and had an Eastern Cooperative Oncology Group performance status (ECOG PS) of 0 or 1. Key exclusion criteria for both trials included having received >1 line of therapy, prior treatment with taxanes (for JVCZ only), and antiangiogenic agents. Full eligibility criteria are provided in the Online Supplementary Methods. All patients provided written informed consent before participation. The protocol was approved by the ethics committee for all participating centers. The study adhered to the Declaration of Helsinki and the International Conference on Harmonization Guidelines for Good Clinical Practice, and applicable local regulations.In JVDB, patients were randomized 1:1:1:1 to receive ramucirumab intravenously in 1 of 4 dosing regimens as shown in Figure S1A: Arm 1, 8 mg/kg (current registered dose) on days 1 and 15 of a 28-day cycle (Q2W) [1]; Arm 2, 12 mg/kg (Q2W); Arm 3, 6 mg/kg weekly (on days 1, 8, 15, and 22 of a 28-day cycle [QW]); Arm 4, 8 mg/kg on days 1 and 8 of a 21-day cycle (D1D8-Q3W). Stratification factors were body weight (<60 kg vs. ≥60 kg) and ECOG PS (0 vs. 1).In JVCZ, patients were randomized 1:1 after ECOG PS (0 vs. 1) stratification to 1 of the 2 dosing regimens: either 8 mg/kg Q2W (the currently registered dose) or 12 mg/kg Q2W ramucirumab, both in combination with paclitaxel at 80 mg/m2 on days 1, 8, and 15 of a 28-day cycle (Figure S1B).Based on PK simulations, the ramucirumab EDRs for both studies were expected to produce higher exposure and potentially better clinical activity outcomes relative to the 8 mg/kg-Q2W dosing regimen.Tumor responses were assessed radiographically according to RECISTv1.1 [11] during study treatment, every 6 weeks (±7 days) for the first 6 months in study JVDB, and every 9 weeks (±7 days) thereafter. In JVCZ, imaging was performed every 8 weeks (±7 days). Treatment continued until disease progression, unacceptable toxicity, or discontinuation for any other reason. Adverse events (AEs) were evaluated throughout the study and for 30 days after treatment discontinuation according to the Common Terminology Criteria for Adverse Events, Version 4.0, and were judged by the investigator as related or unrelated to study treatment. PK and immunogenicity sampling was carried out according to Table S1. Ramucirumab serum concentrations were analyzed using a validated Enzyme-Linked Immunosorbent Assay (ELISA, Intertek Pharmaceutical Services, San Diego, CA, USA). Immunogenicity testing was performed using a validated assay (BioAgilytix Inc., Durham, NC, USA).For JVDB, the primary objective was to evaluate the PK and safety of various dosing regimens of ramucirumab monotherapy. Secondary objectives were PFS rate at the first 6-week tumor assessment (6-week PFS rate) as determined by the investigator per RECISTv1.1, and immunogenicity for ramucirumab. Exploratory objectives were PFS, ORR, and OS.For JVCZ, the primary objective was to compare PFS between ramucirumab 12 mg/kg plus paclitaxel and placebo plus paclitaxel using RAINBOW data (PFS Analysis 1) through inter-trial analysis. The key assumption was that the study populations in JVCZ and RAINBOW were similar with respect to predictive and prognostic factors. Secondary objectives were PFS between ramucirumab 12 versus 8 mg/kg (PFS Analysis 2), PK, safety and tolerability, objective response rate (ORR), disease control rate (DCR), and immunogenicity. The trial was not powered for OS, which was an exploratory objective. Efficacy measures are defined in Online Supplementary Methods.Safety analyses were based on all enrolled patients who received at least 1 dose of ramucirumab (JVDB), or a partial dose of ramucirumab or paclitaxel (JVCZ). Efficacy analyses were based on the intention-to-treat population, defined as all patients randomized to study treatment.For JVDB, a sample size of 160 patients (40 per arm) was needed to provide an adequate estimate of the PFS rate at 6 weeks based on an assumed 6-week PFS rate of about 60% for Arm 1 (based on REGARD study) and 80% for the exploratory doses, with a 50% chance to detect a statistical difference at α = 0.05.For JVCZ, an assumed 2-month increase in median PFS between ramucirumab 12 versus 8 mg/kg (PFS Analysis 2) with an estimated hazard ratio (HR) = 0.667 (HR2) was used to estimate the sample size of 191 PFS events observed in JVCZ study. In the primary efficacy analysis (PFS Analysis 1), comparing the PFS between ramucirumab 12 mg/kg and the placebo arm in RAINBOW, HR was calculated by HR1*HR2 using meta-analysis. HR1 = 0.635 was the observed HR for PFS from RAINBOW and HR2 (ramucirumab 12 vs. 8 mg/kg) was estimated using the Cox proportional hazard model based on the JVCZ data only. Only 64 PFS events were needed to achieve a statistical power of 90% for PFS Analysis 1 and thus the sample size of 191 events as determined by PFS Analysis 2 was also sufficient for the meta-analysis in PFS Analysis 1.In both trials, time-to-event variables for both PFS and OS were estimated using Kaplan–Meier methods with 95% confidence intervals (CIs). All tests of treatment effects were conducted using the log-rank test at a 2-sided alpha-level of 0.05. A Cox proportional hazard model was used to estimate the HR between the treatment arms and the corresponding CIs and Wald p-values. The proportions of patients achieving objective response and disease control between treatment arms were compared using the Cochran–Mantel–Haenzel test. Statistical analyses were performed using SAS software (SAS, Version 9.1.2 or higher, Cary, NC, USA).For study JVDB, between 14 July 2015 and 18 August 2016, 164 of 205 screened patients were randomly assigned to one of the four ramucirumab monotherapy arms (Figure S1A). Of these, 161 patients in total were treated with ramucirumab at the corresponding dosing regimen. As of data cutoff (18 November 2016), 142 (86.6%) of the randomized patients had discontinued, 56.7% due to disease progression.For study JVCZ, between 22 October 2015 and 26 January 2017, 245 of 305 screened patients were randomly assigned to receive ramucirumab plus paclitaxel at different dosing regimens (Figure S1B). Of these, 243 received study treatment. As of data cutoff (27 October 2017), 232 (94.7%) of the randomized patients had discontinued, 65.3% due to disease progression.The baseline characteristics of patients were well balanced across treatment arms in both studies (Table S2). Overall, the majority of patients were male, with measurable disease, and ECOG PS of one; the majority had a diagnosis of gastric adenocarcinoma compared to GEJ adenocarcinoma. The median relative dose intensity was high in both studies, ranging from 94 to 100% for ramucirumab across both studies, and 88.5% for paclitaxel in the JVCZ combination study (Table S3). The median duration of ramucirumab therapy across all treatment arms is shown in Table S3.PK results are presented in Figure 1. Ramucirumab trough concentrations were higher in the 3 EDRs in JVDB (Arms 2–4) than those observed in the standard 8 mg/kg-Q2W regimen (Arm 1). Ramucirumab exposure (peak and trough concentrations) was increased as expected (by ~50%), between 8 and 12 mg/kg (Figure 1A, JVDB and Figure 1B, JVCZ). In JVDB, six patients (6.9%) had one sample positive for anti-ramucirumab antibodies at any time during the study. No patients met the criteria for treatment-emergent antibodies or had neutralizing antibodies. In JVCZ, 18 (9.3%) patients had samples positive for anti-ramucirumab antibodies of the 194 with evaluable samples. In the 12 mg/kg-Q2W ramucirumab arm, two patients met the criteria for treatment-emergent antibodies; one patient met these criteria in the 8 mg/kg-Q2W ramucirumab arm. No neutralizing antibodies were detected.In the JVDB monotherapy study, the 6-week PFS rate ranged from 43.9% in Arm 1 (currently registered 8 mg/kg-Q2W dosing regimen) to 61.9% in Arm 2 (12 mg/kg-Q2W), but it was not significantly different between any of the three EDRs and Arm 1 (Figure 2A). The median PFS and OS were approximately 1 month longer in the ramucirumab 12 mg/kg-Q2W regimen compared with Arm 1 (Figure 2B,C). Post-discontinuation therapies for each arm are summarized in Table S4. No complete or partial responses were observed in the 8 mg/kg-Q2W regimen, while nine patients across the three EDRs had a best overall response of partial response, with the 12 mg/kg-Q2W regimen displaying the higher ORR (9.5%). The number of patients with stable disease was also numerically superior for the three EDRs compared with the 8 mg/kg-Q2W arm (Figure 2A).In the JVCZ study, we observed a statistically significant treatment effect on PFS in the ramucirumab 12 mg/kg plus paclitaxel combination arm when compared to the placebo plus paclitaxel arm of RAINBOW [5] using meta-analysis (PFS Analysis 1, p-value = 0.0035, Figure 3A), which was our primary endpoint. The same was not observed for the exploratory analysis of OS comparison, which was anticipated since study JVCZ was not powered to show a difference in OS (Analysis 1). In addition, median PFS and OS were not significantly different between the 12 mg/kg and the 8 mg/kg arms using JVCZ data alone (Analysis 2, Figure 3A), despite OS being numerically longer in the ramucirumab 12 mg/kg combination arm (9.72 months vs. 7.59 months) (Figure 3C). Median PFS was 5.42 months (95% CI 4.40, 6.01) and 5.16 months (95% CI 3.81, 5.65) in the ramucirumab 12 and 8 mg/kg plus paclitaxel combination arms, respectively (Figure 3B). Analysis of PFS and OS data by ramucirumab serum minimum concentration (Cmin,1) quartiles in our patient population treated with ramucirumab plus paclitaxel (Figure 4) did not show a statistically significant difference in survival in subsets of patients displaying higher ramucirumab serum concentrations, although this was not a powered analysis. Confirmed ORR and DCR were similar between the two JVCZ treatment arms (Table S5).In the JVDB study, treatment-emergent adverse events (TEAEs) occurred in 131 patients (81.4%), of whom 69 patients (42.9%) had TEAEs that were considered related to study treatment (Table S6). The incidence of any grade or grade ≥3 TEAEs was similar among the four treatment arms. More patients reported TEAEs in the ramucirumab 6 mg/kg-QW arm, and more serious AEs (SAEs) were observed in the ramucirumab 8 mg/kg-D1D8-Q3W dosing regimen compared to the 8 mg/kg-Q2W regimen (Table S6). A detailed listing of TEAEs and SAEs reported by treatment arm and preferred term is provided in Table 1. Of note, grade ≥3 fatigue was reported in six patients (14.3%) on the 12 mg/kg-Q2W treatment arm; no other grade ≥3 TEAE was reported in more than 10% of patients on any dosing regimen. Two patients died in the 6 mg/kg-QW arm because of SAEs that were deemed related to ramucirumab: one from gastric hemorrhage and one from respiratory failure (Table 1 and Table S6).AEs of special interest (AESIs) for ramucirumab, regardless of causality, occurred in 69 patients (42.9%) in the JVDB study. The most common AESIs reported in ≥5% of patients were bleeding/hemorrhage events (19.3%, 31 patients), hypertension (11.8%, 19 patients), and liver injury events (8.7%, 14 patients, mostly aspartate aminotransferase increased) and proteinuria (5.6%, 9 patients), Table S7. The 6 mg/kg-QW dosing regimen had the highest percentage of patients with any grade bleeding/hemorrhage events (34.1%, 14 patients), the most common of which was epistaxis, predominantly of grade 1 or 2 severity. There were two fatal events of gastric hemorrhage reported as the primary cause of death.In the JVCZ study, TEAEs occurred in 243 patients (96.3%), with 210 (86.4%) experiencing TEAEs deemed related to study treatment (Table S6). Grade 3 or worse TEAEs occurred in 167 patients (68.7%), with similar incidence between the two treatment arms. A detailed listing of TEAEs and SAEs by treatment arm and preferred term is provided in Table 2. SAEs were reported at a higher rate in the 12 mg/kg ramucirumab arm compared to the 8 mg/kg ramucirumab arm (38.2 vs. 25.8%), and discontinuation of study treatment because of AEs was also higher (18.7 vs. 9.2%) (Table S6). Incidence of death due to AEs while on study treatment or within 30 days of treatment discontinuation was the same for both arms (7.3 vs. 6.7%) (Table S6). A higher number of patients reported the SAE of neutropenia in the 12 mg/kg-Q2W ramucirumab arm compared with the 8 mg/kg-Q2W arm (4.9 vs. 0.8%), but the incidence of febrile neutropenia was comparable between arms (3.3 vs. 2.5%) (Table 2). There was a similar rate of deaths attributed to SAEs related to study treatment while on therapy in both arms: five (4.1%) in the 12 mg/kg-Q2W ramucirumab arm (general physical health deterioration [n = 2], esophageal fistula, shock hemorrhagic, and tumor perforation) and four patients (3.3%) in the 8 mg/kg-Q2W arm (bone marrow failure, pneumonitis, respiratory distress, and upper gastrointestinal hemorrhage).In the JVCZ study, AESIs for ramucirumab occurred regardless of causality in 145 patients (59.7%). The most common AESIs reported in ≥2% of patients were epistaxis, hypertension, and aspartate aminotransferase increase (Table S7). The incidence of most AESIs was similar across treatment arms, except for liver-related events, where a higher number of patients reported any grade and grade ≥3 liver-related events in the 12 mg/kg-Q2W ramucirumab arm versus the 8 mg/kg-Q2W arm (Table S7). This was largely due to alanine aminotransferase and aspartate aminotransferase increase, occurring mostly at grade 1–2 severity.JVDB and JVCZ both examined whether higher doses of ramucirumab as monotherapy or in combination with paclitaxel would improve efficacy while maintaining a favorable safety profile in patients with advanced gastric/GEJ adenocarcinoma. The dosing regimen of ramucirumab used in the 3 EDRs of the JVDB and JVCZ trials differs from the approved dose (8 mg/kg-Q2W). The rationale for these studies was based on exploratory exposure–response analyses of prior trials, suggesting higher exposure might achieve better outcomes [8,9]. For the dosing regimens included in the current studies (8- and 12 mg/kg-Q2W), similar serum ramucirumab concentrations were observed between the JVCZ and JVDB respective dosing regimens. Exposure achieved with the 8 mg/kg-Q2W dosing regimen in JVDB and JVCZ was consistent with that observed in REGARD and RAINBOW phase III studies [4,5]. In JVDB, higher ramucirumab trough concentrations were associated with the three EDRs compared to the currently registered 8 mg/kg-Q2W dosing regimen. In JVCZ, ramucirumab exposure also increased as expected (by ~50%) in the 12 mg/kg-Q2W arm compared to the 8 mg/kg-Q2W regimen.The safety profile of ramucirumab observed in these studies was consistent with the established safety profile reported in REGARD and RAINBOW trials. In JVDB, the incidence of most AE categories was similar across all four dosing regimens, though the occurrence of SAEs was slightly higher in the experimental ramucirumab 8 mg/kg-D1D8-Q3W arm. In addition, the 6 mg/kg-QW dosing regimen had higher incidences of bleeding/hemorrhage events, including fatal gastrointestinal hemorrhage events. This could have been due to the slightly higher ramucirumab trough concentrations at some time points in these experimental arms. Our conclusions on the effect of alternative dose timings on safety are limited by the small sample size and high standard deviation. Nevertheless, the safety profile described here for these ramucirumab-specific AEs is consistent with that described in the meta-analysis of six randomized trials in 4996 patients [10]. In the JVCZ ramucirumab 12 mg/kg-Q2W plus paclitaxel arm, more patients experienced SAEs and discontinuation of study treatment due to AEs compared with the 8 mg/kg-Q2W arm, but the rates of these events are in line with that observed previously [5]. Our results demonstrate that no clinically meaningful additional toxicity was observed with alternative doses of ramucirumab as a single agent or in combination with paclitaxel when compared with the currently registered ramucirumab regimen of 8 mg/kg-Q2W in patients with gastric/GEJ adenocarcinoma.In JVDB, the hazard ratios for the 12 mg/kg-Q2W and 8 mg/kg-D1D8–Q3W dosing regimens suggest a potential trend toward improved PFS and OS compared with the currently registered ramucirumab 8 mg/kg-Q2W dose. However, this study was not powered for statistical comparisons on efficacy parameters due to the small size of the arms, preventing us from making additional inferences. Additional studies would be needed to confirm these results in a higher number of patients.The JVCZ study met its primary objective, demonstrating a statistically significant treatment effect on PFS with ramucirumab 12 mg/kg plus paclitaxel versus the placebo plus paclitaxel arm from RAINBOW. Ramucirumab exposure in the 12 mg/kg regimen was increased by 50% from that achieved with the 8 mg/kg regimen; nevertheless, no statistically significant difference was observed in terms of PFS between these two regimens, which was our secondary objective. The safety data from JVCZ are consistent with that previously reported for RAINBOW [5].Overall, our results demonstrate no statistically significant additional survival benefit associated with a higher dose and exposure of ramucirumab as a single agent or in combination with paclitaxel compared with the current approved dose of 8 mg/kg-Q2W. Furthermore, there was no meaningful additional toxicity, likely because the doses were still far below the MTD of 13 mg/kg/week [2]. The safety profile appeared consistent across EDRs and with the previously observed safety profile of ramucirumab in gastric/GEJ adenocarcinoma. This is despite prior exposure-response analyses suggesting a positive relationship between ramucirumab exposure and survival [8,9], leading us to hypothesize that an improved benefit could be achieved by increasing the dose to 12 mg/kg. A similar lack of clinically relevant exposure-efficacy benefit has been reported for other antineoplastic monoclonal antibodies (mAbs), such as pembrolizumab in melanoma and non-small cell lung cancer [12] and trastuzumab in metastatic gastric adenocarcinoma, where higher concentrations did not translate into increased survival [13]. A pooled analysis of ipilimumab phase II and III data in melanoma also showed no benefit with different dosing [14] despite initial results suggesting an improvement in OS at the expense of safety [15,16]. The underlying mechanisms for this lack of dose dependency are unclear but several hypotheses have been suggested, including a correlation between cachexia and mAb catabolism [12]. On the other hand, a dose-response effect translating into higher efficacy with a higher dose of bevacizumab has been reported in a randomized phase II trial of bevacizumab in combination with chemotherapy for non-small cell lung cancer [17].A population PK model for ramucirumab has been developed [18], and none of the clinical factors known to be prognostic for response in gastric cancer (e.g., ECOG status, presence of metastatic sites) were found to be significantly correlated to PK parameters or exposure. Consequently, the lack of an exposure-efficacy relationship for study JVCZ is unlikely to be biased by a confounding relationship between efficacy prognostic factors and PK. The exposure-response analysis presented here did not establish that increasing the dose from 8 to 12 mg/kg improved efficacy.Given the lack of a statistically significant greater benefit seen with an increase of up to 50% in doses of ramucirumab, both as monotherapy or in combination with paclitaxel in the JVDB and JVCZ studies, we conclude the approved intravenous ramucirumab dosing regimen of 8 mg/kg Q2W is clinically appropriate for the second-line treatment of adult patients with advanced or metastatic gastric/GEJ adenocarcinoma.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051168/s1, Online Supplementary Methods S1: Full inclusion/exclusion criteria for studies JVDB and JVCZ, Online Supplementary Methods S2: Efficacy measures for studies JVDB and JVCZ; Figure S1: CONSORT diagrams. (A) Study I4T-MC-JVDB (B) Study I4T-MC-JVCZ; Table S1: Ramucirumab dosing and pharmacokinetic and immunogenicity sampling schedule, Table S2: Combined demographics and disease characteristics of patients randomized to ramucirumab monotherapy (JVDB) or combination therapy (JVCZ), Table S3: Summary of drug exposure (safety population) for studies JVDB and JVCZ, Table S4: Summary of post-treatment discontinuation therapies received by patients in studies JVDB and JVCZ (by arm), Table S5: Best overall response for JVCZ study per RECISTv1.1 in solid tumors, Table S6: Safety overview (JVDB/JVCZ), Table S7: Treatment-emergent AESIs for ramucirumab (regardless of causality) reported in the safety population of (A) ramucirumab monotherapy JVDB, and (B) combination therapy JVCZ studies.Conceptualization R.W., D.F., L.G. and J.M.O.; validation R.W.; formal analysis R.W., D.F., L.G. and J.M.O.; investigation M.A.S., A.A.U., I.B., W.M., R.G.S., T.S. (Tomasz Sarosiek), S.B., M.S., C.G.-M., C.M., M.Ö., J.P., H.P.K., E.W., T.B., D.S., I.C., M.J., I.V., A.V.L., P.C.E., T.S. (Tomas Salek), C.P., C.T., E.M. and J.A.A.; writing—original draft preparation R.W., D.F., L.G. and J.M.O.; writing—review and editing M.A.S., A.A.U., I.B., W.M., R.G.S., T.S. (Tomasz Sarosiek), S.B., M.S., C.G.-M., C.M., M.Ö., J.P., H.P.K., E.W., T.B., D.S., I.C., M.J., I.V., A.V.L., P.C.E., T.S. (Tomas Salek), C.P., C.T., E.M., R.W., D.F., L.G., J.M.O. and J.A.A.; visualization R.W. and L.G.; supervision, M.A.S., D.F. and J.A.A. All authors have read and agreed to the published version of the manuscript.This work was funded by Eli Lilly and Company.The studies were conducted in accordance with the International Conference on Harmonization Good Clinical Practice guidelines, the Declaration of Helsinki, applicable local regulations, and were approved by each institution’s ethical review board and registered at Clinicaltrials.gov (registration numbers NCT02443883 and NCT02514551).Informed consent was obtained from all subjects involved in the study.Eli Lilly and Company provides access to all individual participant data collected during the trial, after anonymization, with the exception of pharmacokinetic or genetic data. Data are available to request 6 months after the indication studied has been approved in the USA and EU and after primary publication acceptance, whichever is later. No expiration date of data requests is currently set once data are made available. Access is provided after a proposal has been approved by an independent review committee identified for this purpose and after receipt of a signed data sharing agreement. Data and documents, including the study protocol, statistical analysis plan, clinical study report, blank or annotated case report forms, will be provided in a secure data sharing environment. For details on submitting a request, see the instructions provided at www.vivli.org (accessed on 22 December 2021).We thank all the patients and their caregivers for participation in this trial. We also thank all the investigators and their support staff who generously participated in this work. We wish to thank Francesca Russo for contributions to this study and Sophie Callies for consulting on the pharmacokinetic analyses. Medical writing support, funded by Eli Lilly and Company, was provided by Erika Wittchen and Anchal Sood, and editorial support by Dana Schamberger and Antonia Baldo, all from Syneos Health.M.A.S. reports grants from Eli Lilly and Company. A.A.U. reports travel and accommodation fees from Amgen, Angelini, Janssen, Merck Sharp & Dohme, Pfizer, and Roche. M.S. reports grants for clinical trials from Eli Lilly and Company, Roche, MSD, Merck, Astra Zeneca, Tesaro, Regeneron Pharmaceuticals, Astellas, Novartis, Bristol Myers Squibb, Pfizer, AbbVie, GlaxoSmithKline, Mylan, Samsung Pharmaceuticals, Bioven, BeiGene, PharmaMar, Clovis, Bayer, Gilead, and Amgen. C.G.-M. reports personal fees from Eli Lilly and Company, Roche, BMS, and Eisai-MSD. H.K. reports research funding from Roche, Pfizer, Novartis, Amgen, Bayer, Genesis, Eli Lilly and Company, MSD, Janssen Pharmaceuticals, and Merck-Serono; consulting/advisory fee from Roche, Novartis, MSD, Genesis, Pfizer, Eli Lilly and Company, LEO Pharma, Amgen, and Merck-Serono; and travel/accommodations fees from Roche, Novartis, Enorasis, and Pfizer. TB reports personal fees and institutional payments for clinical studies from Eli Lilly and Company, Roche, Bristol Myers Squibb, Bayer, Merck, Eisai, Exelixis, AstraZeneca, and Sanofi; personal fees from Astellas, Janssen, Ipsen, and Servier; non-financial support from Bristol Myers Squibb and Ipsen; grants from Servier. P.E. reports honoraria and advisory/consultancy fees from ALX Oncology, Arcus Biosciences, Astellas, AstraZeneca, Blueprint Medicines, Bristol Myers Squibb, Celgene, Daiichi Sankyo, Five Prime Therapeutics, Ideaya Biosciences, Istari Oncology, Legend Biotech, Eli Lilly and Company, Loxo Oncology, Merck, Ono Pharmaceutical, Taiho Pharmaceutical, Takeda Pharmaceutical Company, Turning Point Therapeutics, Xencor, and Zymeworks. L.G. is a former employee of Eli Lilly and Company. R.W., D.F., L.G., and J.M.O. are full-time employees of Eli Lilly and Company. All other authors have declared no conflicts of interest.(A) JVDB ramucirumab trough and peak serum concentrations (mean ± SD) before and after administration of indicated doses of ramucirumab monotherapy (n = 160 patients). (B) JVCZ ramucirumab trough and peak serum concentrations before and after administration of ramucirumab 8 or 12 mg/kg in combination with paclitaxel (n = 232 patients). Abbreviations: Conc—concentration; D—day; Q2W—every 2 weeks; Q3W—every 3 weeks; QW—weekly; SD—standard deviation; w—weeks.Six-week PFS and objective responses (A), PFS (B), and OS (C) for the standard regimen (Arm 1) and the 3 EDRs (Arms 2, 3 and 4) of ramucirumab monotherapy (JVDB). Abbreviations: EDRs—exploratory dosing regimens; OS—overall survival; PFS—progression-free survival.PFS and OS analysis of ramucirumab 12 mg/kg plus paclitaxel in JVCZ versus placebo plus paclitaxel in RAINBOW (Analysis 1) and 12 mg/kg versus 8 mg/kg within JVCZ (Analysis 2) (A), PFS (B), and OS (C). Abbreviations: OS—overall survival; PFS—progression-free survival.PFS (A) and OS (B) of JVCZ safety population by ramucirumab Cmin,1 quartiles. Abbreviations: Cmin,1—serum minimum concentration; OS—overall survival; PFS—progression-free survival.Study JVDB: treatment-emergent adverse events and serious adverse events.Abbreviations: D1D8-Q3W—day 1 and 8 of a 3-week cycle; MedDRA—Medical Dictionary for Regulatory Activities Version 19.1; N = number of patients in the safety population; Q2W—every 2 weeks; Q3W—every 3 weeks; QW—weekly; SAE—serious adverse event; TEAE—treatment-emergent adverse event. a Ordered by decreasing frequency in Arm 1. Italicized items are consolidated terms incorporating the multiple MedDRA preferred terms. b One patient died in the 6 mg/kg QW arm because of the SAE gastric hemorrhage, deemed related to ramucirumab. c One patient died in the 6 mg/kg QW arm because of the SAE respiratory failure, deemed related to ramucirumab.Study JVCZ: treatment-emergent adverse events and serious adverse events.Abbreviations: MedDRA—Medical Dictionary for Regulatory Activities Version 20.1; N—number of patients in the safety population; SAE—serious adverse event; TEAE—treatment-emergent adverse event. a Italicized items are consolidated terms incorporating the multiple MedDRA preferred terms. b Ramucirumab was administered on Days 1 and 15 of each 28-day cycle. Paclitaxel was administered on Days 1, 8, and 15 of each 28-day cycle. c As used in this table, “abdominal pain” includes the MedDRA preferred terms of “abdominal pain”, “abdominal pain upper”, as well as “gastrointestinal pain” and “abdominal pain lower”.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01169.txt

@@ -0,0 +1 @@
+Imaging plays a crucial role in the accurate staging of prostate cancer. Prostate-specific membrane antigen (PSMA) is overexpressed in prostate cancer cells, and targeting the PSMA protein for diagnostic purposes has become of great clinical value. Another valuable feature of PSMA is its opportunity to serve as a target for delivering radionuclide therapy to cancer cells. PSMA-ligands can be labeled with various radionuclides, such as alpha and beta-emitters. This review offers an overview of the literature on recent developments in nuclear medicine regarding PSMA in prostate cancer diagnostics and targeted radionuclide therapy.Targeting the prostate-specific membrane antigen (PSMA) protein has become of great clinical value in prostate cancer (PCa) care. PSMA positron emission tomography/computed tomography (PET/CT) is increasingly used in initial staging and restaging at biochemical recurrence in patients with PCa, where it has shown superior detection rates compared to previous imaging modalities. Apart from targeting PSMA for diagnostic purposes, there is a growing interest in developing ligands to target the PSMA-protein for radioligand therapy (RLT). PSMA-based RLT is a novel treatment that couples a PSMA-antibody to (alpha or beta-emitting) radionuclide, such as Lutetium-177 (177Lu), to deliver high radiation doses to tumor cells locally. Treatment with 177Lu-PSMA RLT has demonstrated a superior overall survival rate within randomized clinical trials as compared to routine clinical care in patients with metastatic castration-resistant prostate cancer (mCRPC). The current review provides an overview of the literature regarding recent developments in nuclear medicine related to PSMA-targeted PET imaging and Theranostics.Prostate cancer (PCa) is the second-most common malignancy worldwide, and it is the fifth leading cause of cancer-related mortality among men [1]. When detected at an early stage, patients tend to have an excellent prognosis. However, the course of PCa is heterogeneous and varies from indolent to highly aggressive disease [2,3]. Therefore, accurate staging and risk stratification are essential in the management of patients with PCa, given the wide variety of therapeutic options that may differ per disease stage. Currently, imaging plays a pivotal role in assessing the disease extent, particularly through targeting the prostate-specific membrane antigen (PSMA) [4]. PSMA is a transmembrane glycoprotein substantially overexpressed in malignant prostate cells [5]. As a result, PSMA is an attractive target for molecular imaging with positron emission tomography (PET) using one of several available radiolabeled PSMA-ligands. However, the expression of PSMA is not restricted to prostate (cancer) cells only and may be found in several non-prostatic diseases [5,6]. In clinical practice, the main indications to perform PSMA PET/computed tomography (CT) are initial staging and restaging at the biochemical recurrence of disease after treatment with curative intent [3,7,8]. Recently, the E-PSMA reporting guidelines have been proposed in order to harmonize protocols and to standardize PSMA PET/CT imaging reporting in PCa [9].Aside from targeting the PSMA protein for diagnostic purposes, there is an increasing interest in using PSMA-radioligands for therapeutic purposes. This approach is called radioligand therapy (RLT). PSMA-RLT combines PSMA-ligands and therapeutic radionuclides to deliver targeted high radiation doses to cancer cells, leading to cellular death. PSMA-ligands can be labeled with either alpha (e.g., Actinium-225 (225Ac), Lead-212) or beta-emitting radionuclides (e.g., Lutetium-177 (177Lu)), with both having different characteristics in terms of physics and radiobiology [10]. Most experience has been gained with 177Lu-PSMA-617 in patients with metastatic castration-resistant prostate cancer (mCRPC) [11,12]. In the VISION trial, treatment with 177Lu-PSMA-617 resulted in an overall survival (OS) benefit of 4 months compared to routine clinical care [12]. This review summarizes the current literature on the recent developments in nuclear medicine regarding PSMA in PCa diagnostics and targeted radionuclide therapy. PCa suspicion rises with an abnormal digital rectal examination (DRE), an elevated serum prostate-specific antigen (PSA)-value, or both. However, PSA is organ-specific, not PCa specific, and might be increased in patients with benign diseases (i.e., prostatitis or benign prostate hyperplasia). Consequently, histopathological tissue assessment of prostate biopsies is required to confirm the diagnosis and estimate its aggressiveness, classified using the Gleason score (GS) [13]. Nevertheless, prostate biopsies are vulnerable to sampling errors, leading to false-negative outcomes and potentially inaccurate tumor evaluation [14]. Therefore, current international guidelines recommend multiparametric magnetic resonance imaging (mpMRI) in patients with an elevated PSA before prostate biopsy, allowing the targeted biopsy of suspicious radiological lesions [3]. Additionally, MRI provides essential information for local staging and planning of curative treatment, such as radical prostatectomy or radiation therapy [3]. Recently, the use of PSMA PET/CT for the initial staging of patients with high-risk PCa has also been recognized based on the results of several prospective studies [15,16].According to the International Society of Urological Pathology (ISUP) 2014, grading systems based on the GS, prostate biopsies are classified into five different grades groups of malignancy, ranging from 1 to 5 [13]. Alongside, the Tumor-Node-Metastasis (TNM) classification system is utilized for the uniform staging of PCa [17]. PCa is classified as an organ-confined (T1 and T2) or locally advanced disease (T3 and T4), the latter indicating that the tumor extends beyond the prostate and may invade adjacent structures. These clinical parameters (i.e., TNM stage, PSA, and ISUP grade) are implemented in the European Association of Urology (EAU) PCa risk categories, dividing patients into low, intermediate, or high-risk disease groups [3]. Higher risk groups are associated with an increased risk of having or developing metastatic disease. This underlines the essence of correct and complete staging in these patients, including assessment of metastatic dissemination. The assessment of regional lymph node metastases (N-status) and distant metastases (M-status) is crucial for the accurate staging of patients with PCa since it affects therapy planning and prognosis. Unfortunately, the median survival of men with newly diagnosed metastatic (M1) PCa is approximately 42 months [18]. Common metastatic sites are local and/or distal lymph nodes and bone, while visceral metastases occur less frequently. According to the EAU guidelines, metastasis screening at initial diagnosis is recommended in intermediate and high-risk disease by at least abdominopelvic imaging and bone scintigraphy (BS) [3]. However, the diagnostic accuracy of these conventional imaging modalities is limited for detecting PCa lesions [19,20]. For example, the sensitivity of CT and MRI for pelvic lymph node detection is only 42% and 39%, respectively [19]. A potential explanation may be that these imaging modalities primarily rely upon lesion morphology (i.e., the shape and size of a lesion) for detection, which might be inaccurate in (early) metastatic PCa with small metastases being missed.Radiolabeled PSMA-ligands have recently been introduced to the rapidly evolving nuclear imaging field. While most studies have investigated its performance in either primary staging or restaging at biochemical recurrence (e.g., rising PSA after local therapy), there is increasing data regarding its use in the follow-up of patients with mCRPC. PSMA-ligands can be labeled with 68Gallium (68Ga) or 18Fluoride (18F). 18F-labeled tracers have increased positron yield and shorter positron range compared with 68Ga-labeled tracers, resulting in a higher resolution of the images, with potentially enhanced detection of (small) metastases. Additionally, 18F has the advantages of a longer half-life (110 versus 68 min for 68Ga), enabling centralized production on a larger scale [21]. 68Ga-PSMA-11 and 18F-DCFPyL are the most commonly used radioligands and are primarily excreted by the urinary tract, often making the interpretation of the prostate bed and/or metastases adjacent to the ureters challenging [4,22]. A relatively novel introduced 18F-labeled tracer is 18F-PSMA-1007, with a comparable diagnostic accuracy as 68Ga-PSMA-11 and 18F-DCFPyL for detecting the local recurrence of PCa in the prostatic fossa [23,24,25,26]. The 18F-PSMA-1007 excretion pathway is mainly by the hepatobiliary tract and marginally by urinary excretion, yielding the potential benefit to differentiate nodal metastases or local recurrence from physiological urinary activity [23,24,27]. A disadvantage of 18F-PSMA-1007 is its high unspecific bone uptake, leading to a greater prevalence of positive PSMA findings attributed to a benign origin. Therefore, extensive reader training is necessary to become familiar with the interpretation and reporting [25,26]. Implementing the recently developed E-PSMA criteria might mitigate these clinically relevant interpretation differences among readers in routine daily practice [9].Recent studies have demonstrated the advantages of PSMA PET/CT in the primary staging of men with PCa compared to conventional imaging modalities [8,16,19,20,28]. For example, Pienta et al. evaluated the performance of 18F-DCFPyL, a second-generation PSMA-ligand PET/CT, in detecting metastatic disease at initial staging in high-risk PCa compared with histopathology in the OSPREY trial. In this prospective multicenter phase II/III trial, a total of 252 patients with high-risk PCa planned for radical prostatectomy with lymph node dissection were included. 18F-DCFPyL PET/CT compared to CT or MRI alone showed higher specificity (97.9% versus 65.1%, respectively), positive predictive value (PPV) (86.7% versus 28.3%, respectively), and negative predictive value (NPV) (83.2% versus 77.8%, respectively), with similar sensitivity (40.3% versus 42.6%, respectively) for the detection of pelvic lymph node involvement (LNI) [8]. Similar results were found when investigating the diagnostic accuracy of 68Ga-PSMA and 18F-DCFPyL PET/CT for lymph-node staging in primary PCa [29,30]. The prospective cohort study by van Kalmthout et al. reported a limited sensitivity (41.5%) and high specificity (90.9%) for detecting pelvic lymph node metastases with 68Ga-PSMA PET/CT in patients with newly diagnosed PCa [30]. A similar study from Jansen et al. reported a sensitivity and specificity of 41.2% and 94.0%, respectively, for detecting lymph node metastases with 18F-DCFPyL PET/CT [29]. Nevertheless, mainly based on the encouraging results from the ‘proPSMA’ trial, the European Association of Urology (EAU) guidelines have recently incorporated PSMA PET/CT for initial staging purposes [3]. In this prospective multi-center study, 302 patients with high-risk PCa, prior to curative-intent surgery or radiotherapy, were randomly assigned to conventional imaging with CT and bone scintigraphy or 68Ga-PSMA-11 PET/CT. The accuracy of 68Ga-PSMA PET/CT was 27% higher than that of CT and bone scintigraphy (92% versus 65%; p < 0.0001). Conventional imaging had a lower sensitivity (38% versus 85%) and specificity (91% versus 98%) than PSMA PET/CT. Moreover, the 68Ga-PSMA PET/CT scan induced management change more frequently than conventional imaging, with less equivocal findings and lower radiation exposure [16].A PSMA PET/CT limitation is that a negative PSMA PET/CT cannot rule out lymph node metastases [8,29,30,31]. Consequently, the ePLND remains the gold standard for primary nodal staging, despite known potential complications, such as lymphocele, lymphedema, and deep venous thrombosis [3].In 5–20% of the patients treated with radical prostatectomy (RP), the PSA level remains measurable after treatment [32,33]. Biochemical persistence (BCP) is defined as a detectable PSA level of ≥0.1 ng/mL within 4–6 after RP [34]. Causes of BCP are the presence of (micro)metastases or residual disease in the prostatic tissue. Unfortunately, BCP is associated with more advanced PCa, such as higher pathological tumor stages, higher ISUP grade, positive surgical margins, and an impaired prognosis [33,35,36]. Schmidt-Hegemann et al. more frequently observed pelvic LNI on 68Ga-PSMA PET/CT in patients with BCP than patients who develop biochemical recurrence [37]. The multicenter retrospective study by Farolfi et al. reported that 68Ga-PSMA PET/CT localized PCa in two-thirds of the patients with BCP [38]. Additionally, Meijer et al. analyzed the findings of 68Ga-PSMA PET/CT and 18F-DCFPyL PET in 150 patients with BCP after surgical treatment. They found PSMA positive lesions outside the prostatic fossa in 67% of the patients and in 26% of patients outside the pelvis [39]. Therefore, accurate localization of residual disease with PSMA PET/CT is critical to determine and guide the most effective treatment. PSMA PET/CT has been extensively evaluated in patients with biochemically recurrent disease (BCR) after definite treatment. BCR is defined as a serum PSA of ≥0.2 ng/mL after radical prostatectomy or a serum PSA ≥ 2.0 ng/mL above the nadir after radiation therapy [40,41]. In patients with BCR, identifying the recurrence site is crucial as it directly influences therapeutic decision-making. The detection of metastatic disease is strongly associated with the level of PSA-values when performing the PSMA PET/CT [7,28,42]. Interestingly, Jansen et al. analyzed PSMA PET/CT performed in 63 patients with low PSA levels (<2.0 ng/mL, not meeting BCR criteria) after curative radiotherapy and found PSMA positive lesions in 53/63 patients (84.1%) defined as local recurrence (21 patients) or metastatic disease (32 patients) [43]. Perera et al. reported sensitivities for 68Ga-PSMA PET/CT in detecting BCR of 33%, 45%, 59%, 75%, and 95% for PSA ranges of <0.2, 0.2–0.49, 0.5–0.99, 1.0–1.99, and ≥2.0 ng/mL, respectively [28]. Before the introduction of PSMA PET, prostate cancer molecular imaging was commonly performed using radiolabeled choline-ligands (e.g., 11C-choline and 18F-choline) and more recently 18F-Fluciclovine [15,44,45,46,47]. In the literature, 68Ga-PSMA PET/CT has demonstrated higher detection rates than 11C-Choline PET/CT in BCR, especially in patients with low PSA levels [44,45,46,47,48]. A recent prospective trial by Calais et al. enrolled 50 patients with BCR after RP with low a PSA level (<2.0 ng/mL) to compare the detection rate and reproducibility of 68Ga-PSMA PET/CT versus 18F-Fluciclovine. They found significantly higher detection rates with 68Ga-PSMA PET/CT compared to 18F-Fluciclovine (56% versus 26%; OR 4.8 95%CI: 1.6–19.2, p = 0.0026), also when stratified by PSA level (PSA < 0.5 ng/mL: 46% versus 27%; PSA 0.5–1.00 ng/mL: 67% versus 28%; PSA 1.01–2.00: 67% versus 17%, respectively) [15]. Furthermore, the recent prospective, phase III CONDOR trial by Morris et al. assessed the diagnostic performance of 18F-DCFPyL in patients with BCR with negative or equivocal findings on PET/CT (18F-Fluciclovine or 11C-Choline) or conventional imaging (CT, MRI, or BS). Improved detection rates were found when PSA levels were higher (PSA < 0.5 ng/mL: 36.2%; PSA 0.5–0.99 ng/mL: 51.4%; PSA 1.0–1.99: 66.7%). A high correct localization rate (84.8–87.0% lower bound of 95%CI: 77.8–80.4) was found. Furthermore, disease management was changed in nearly two-thirds of the analyzed patients (63.9%, n = 131) [7]. Apart for cohort A of the phase 2/3 OSPREY trial, cohort B included patients with suspected locoregional recurrence and/or distant metastatic disease on conventional imaging (CT, MRI, or BS). Among all patients, high median sensitivity (95.8%) and PPV (81.9%) of 18F-DCFPyL PET/CT were found for detecting recurrence or metastatic disease, respectively. Moreover, metastatic disease was described in 57.6% of the patients previously staged with locoregional disease on conventional imaging. The sensitivity ranged from 88.9% to 100% and the PPV from 61.5% to 88.9% in patients with low PSA levels (<2.0 ng/mL) [8]. Considering these superior detection rates of PSMA PET/CT on biochemical recurrence of disease, PSMA PET/CT has become the recommended imaging modality for BCR following previous curative-intent therapy (Figure 1) [3]. PSMA PET/CT is increasingly used to select the optimal treatment strategy in patients with BCR, and PSMA PET findings frequently result in management changes [49,50,51]. For example, Meijer et al. found a change of preferred management in 40.7% of the patients with BCR who underwent 18F-DCFPyL PET/CT for restaging after curative-intent treatment [50]. Likewise, Calais et al. assessed the impact of 68Ga-PSMA PET/CT on the treatment plan of BCR and showed a change of management in 53% of the patients [49]. When PCa recurrence is restricted to the prostatic fossa, salvage radiation therapy (SRT) may be considered as a potentially curative treatment option and proves to be the most effective at a PSA value of ≤0.5 ng/mL [34]. However, the findings on PSMA PET/CT before SRT impact the planned treatment by extending the target volume, implying dose escalations, or refraining from radiotherapy [52,53,54]. Since the introduction of PSMA PET/CT, patients with BCR may be diagnosed as having metastatic disease at an earlier stage, also known as ‘stage migration’. Patients with the oligometastatic disease have a limited number of metastases (usually defined as 1–5 metastatic lesions). Metastasis-directed radiotherapy (MDT) on these lesions may postpone the initiation of systemic treatment [55,56,57]. A phase II randomized clinical trial by Philips et al. compared stereotactic body radiation therapy (SBRT) observation in patients with oligometastatic recurrent PCa (up to three metastases) on conventional imaging. 18F-DCFPyL PET/CT was performed at baseline in the patients receiving SBRT, and these results were blinded to the investigative team during therapy planning. A higher number of patients progressed at six months in the observational cohort than into the group allocated to SBRT (61% versus 19%). The SBRT treatment plan was compared to the results of the PSMA PET/CT, and patients were divided into a total and subtotal consolidation of PSMA avid lesions. Total consolidation of PSMA lesions decreased the risk of new lesions at six months (16% versus 63%) [57]. This study highlights the impact of PSMA PET/CT in planning MDT in patients with oligometastatic disease. However, the long-term effect on overall survival and quality of life are still to be demonstrated. Castration-resistant prostate cancer (CRPC) is defined as biochemical or radiological progression of disease on conventional imaging in the presence of castration levels of serum testosterone (i.e., <50 ng/dL) [34,58]. In CRPC, the number of available therapeutic choices has increased, while the optimal treatment strategy is not fully established [34,59,60,61,62,63]. Current guidelines (PCWG3 and EAU) recommend conventional imaging in combination with regular blood tests for staging and evaluating disease progression in mCRPC patients, but their sensitivity is known to be limited (Figure 2) [34,58]. For example, the multicenter retrospective study of Fendler et al. was designed to assess 68Ga-PSMA PET performance in CRPC patients without metastases on conventional imaging. Distant metastatic disease was found in 55% of the included patients [64]. More sensitive detection with PSMA PET, and potentially earlier detection of metastatic disease, could impact the course of the disease and may facilitate the initiation of early treatment or timely therapy switch to another therapy [65]. However, the resulting improvement in oncological outcomes has not yet been demonstrated.PSMA PET/CT could be performed for selecting patients for PSMA-directed RLT and (re)staging during or after treatment. It is essential to assess the level of PSMA expression before initiating RLT, as PSMA expression in mCRPC disease is known to be highly variable both within and between patients [66]. As a consequence, approximately one-third of the patients will not respond to PSMA-RLT. Hence, identifying predictors of treatment response could be of great value [67]. Ferdinandus et al. described that a higher platelet level and need for pain medication were significant predictors of a poor treatment response to 177Lu-PSMA-617, and PSMA expression on 68Ga-PSMA PET/CT did not predict PSA response [68]. In a similar cohort, Emmett et al. aimed to identify predictors of treatment response in mCRPC patients treated with 177Lu-PSMA-617. They found a strong correlation of PSMA expression (standardized uptake value (SUV): SUVmax and SUVmean) on 68Ga-PSMA PET/CT at baseline imaging with a treatment response of more than 30%. The location or volume of metastases were no predictors of treatment response [69]. In recent years, a variety of reporting systems have been provided, including staging and lesion characterization, to improve consistent PSMA PET/CT describing [70,71]. Furthermore, the newly proposed E-PSMA consensus guidelines, endorsed by the European Association of Nuclear medicine, offers PSMA PET/CT interpretations and reporting statements to create more uniform and standardized reports for clinical use [72]. These guidelines incorporate earlier proposed PSMA-RADS (PSMA-reporting and Data system) and PROMISE (Prostate Cancer Molecular Imaging Standardized Evaluation) criteria. The PSMA-RADS categorizes PSMA PET/CT findings into five categories based on the probability of malignancy [71]. Furthermore, the PROMISE criteria include the intensity of PSMA expression (ranging from 0–3) and the molecular imaging TNM scores (miTNM score) [70]. Recently, a deep learning algorithm (aPROMISE) has been developed for the automated analysis of PSMA PET images to provide a consistent and standardized evaluation. However, the results of the aPROMISE technology require further validation before it can be translated into clinical practice [73].Aside from targeting PSMA for diagnostic purposes, another valuable feature of PSMA is its opportunity to serve as a target for delivering radionuclides (therapeutic agents) to cancer cells. Using the same target for diagnosis and therapeutics is referred to as Theranostics. Recently, novel radionuclides have been developed and proposed to be used as RLT in clinical practice for PCa management. For example, PSMA-ligands can be labeled with varying radionuclides, such as alpha and beta-emitters [10]. The most frequently used radionuclides for PSMA-RLT are Lutetium-177 (177Lu), which decays by beta-emission, and Actinium-225 (225Ac), alpha-emission. There are several clinically relevant differences between alpha and beta-particles (Table 1) [10]. Alpha-particles have a larger mass and carry higher energies. Alpha-particles have high linear-energy transfer (LET), defined as the amount of energy a particle can transmit along its track. This leads to more damage down their track and causes irreparable double-strand DNA breaks in tumor cells. Alpha-particles have a limited range in tissue (0.05–0.08 mm), providing more controlled and selective irradiation of cancer cells with minimal impact on neighboring tissue [10,74,75]. In contrast, beta-particles have a small mass and a more extended range in tissue (0.62 mm). However, they have less energy in comparison with alpha particles. The LET produced by beta-particles is relatively low, resulting in single-strand DNA breaks, which are repairable and thus may be less effective in damaging PCa cells [10,75]. However, the advantage of the beta-emitter, 177Lu-PSMA is its favorable toxicity profile with less severe side-effects. PSMA-617 is the most commonly used ligand in RLT, which can be coupled to Lutetium-177, resulting in 177Lu-PSMA-617 [76]. In addition, 177Lu can also be attached to the PSMA Imaging and Therapy ligand (177Lu-PSMA I&T) [77]. However, the use of 177Lu-PSMA-617 might be preferred in clinical practice compared to 177Lu-PSMA I&T, possibly due to reduced uptake in the kidney [78]. RLT with 177Lu-PSMA has mainly been studied in mCRPC, showing promising results as a potential treatment approach with a low toxicity profile [11,12,67,79,80,81,82,83]. Several retrospective studies have outlined the biochemical (PSA) response of 177Lu-PSMA-617 in mCRPC (see also Table S1) [84,85,86,87,88,89,90]. Kratochwil et al. reported any PSA response from baseline in 21 (70%) of 30 patients, and a PSA decline of more than 50% was found in 43% (13/30) after 177Lu-PSMA-617 treatment [88]. Similarly, in a study including 100 mCRPC patients with a history of treatment with enzalutamide or abiraterone, Ahmadzadehfar et al. reported any PSA decline and a PSA decline of >50% in 69% and 38% after 177Lu-PSMA-617 therapy [84]. In another study, Ahmadzadehfar et al. evaluated the patient response to the second and third cycle of 177Lu-PSMA-617 in 52 patients and found PSA decline > 50% in 60% of the patients [85]. In a retrospective study of Brauer et al., any PSA decline was found in 91% of the patients (n = 45), and a PSA reduction of greater than 50% occurred in 53%. Any PSA decline after the first treatment cycle was significantly associated with a longer OS [86]. Rahbar et al. included patients with mCRPC treated with 177Lu-PSMA-617 to assess the efficacy and safety of 177Lu-PSMA-617. A PSA decline of 50% or more was found in 45% of the patients. Grades 3 and 4 hematotoxicity occurred in 12% of the patients, and xerostomia was reported in 8% [89]. Another recent publication on 177Lu-PSMA-617 conducted by Rahbar et al. recorded any PSA response in 67% of the 104 included men and a PSA decline of >50% in 33%. Any PSA decline after the first cycle was associated with a longer OS than PSA progression (62.9 versus 47.0 weeks). A PSA decline greater than 50% was not prognostic for overall survival [90].Hofman et al. conducted a single-center, phase II trial including mCRPC patients with progressive disease after conventional treatment. Treatment with 177Lu-PSMA-617 treatment resulted in any PSA level decline in 97% of the patients and a PSA decline of ≥50% in 57%. Most registered adverse events (AE) were xerostomia grade I (87%), transient nausea (50%), and fatigue grade I–II (50%). Grade 3–4 thrombocytopenia due to 177Lu-PSMA-617 occurred in 13% of the patients [80].The randomized, multicenter, phase II TheraP trial compared 177Lu-PSMA-617 (up to six cycles every six weeks) to cabazitaxel (up to 10 cycles every three weeks) in 200 patients with progressive post-docetaxel mCRPC. Patients treated with 177Lu-PSMA-617 showed a ≥50% PSA response more frequently than patients treated with cabazitaxel (66% versus 37%, p < 0.0001). In addition, fewer grade III and IV AE were observed in patients who underwent 177Lu-PSMA-617 treatment (33% versus 53%) [11]. Furthermore, the randomized, phase III VISION trial by Sartor et al. assessed 831 patients with mCRPC diagnosed with at least one positive lesion on 68Ga-PSMA-11 PET/CT. The patients previously underwent treatment with minimal one androgen receptor signaling pathway inhibitor and taxane chemotherapy. The patients were randomized 2:1 to receive 177Lu-PSMA-617 (every six weeks up to four–six cycles) plus standard of care (SOC; (n = 551) or SOC alone (n = 280). The median imaging-based progression-free survival was improved by 5.3 months in the 177Lu-PSMA-617 group compared to the control group (8.7 versus 3.4 months, respectively; p < 0.001). In addition, there was a significant median OS benefit in favor of 177Lu-PSMA-617 (15.3 versus 11.3 months, respectively; p < 0.001). As expected, treatment with 177Lu-PSMA-617 led to a higher incidence of grade 3 AE, or higher, than the control group (52% versus 38%). The most-reported AE were fatigue, dry mouth, and nausea grade I or II. Nevertheless, a low incidence of AE led to alternation of the doses or discontinuation of the study, and treatment with 177Lu-PSMA-617 was considered safe [12]. Challenges remain in the prediction of treatment response and survival in 177Lu-PSMA therapy. In several studies, (changes in) metrics quantifying the burden of PSMA-positive disease on PET were associated with treatment response and survival to 177Lu-PSMA radioligand therapy in patients with mCRPC [91,92,93] There is increasing interest in positioning PSMA-radioligand therapy in the (earlier) hormone-sensitive stage. It is hypothesized that in metastatic hormone-sensitive prostate cancer (mHSPCa), the initiation of androgen deprivation therapy (ADT) can be deferred, and, ultimately, the OS could be improved. Several studies are ongoing in patients with mHSPCa, and results are eagerly awaited [NCT04443062; NCT04343885; NCT04720157]. The most commonly used alpha-emitter for PSMA-ligand treatment is 225Ac-PSMA-617 (see also Table S2). A retrospective study by Kratochwil et al. included 40 patients with mCRPC who underwent treatment with 225Ac-PSMA-617 (every two months up to three cycles). In total, 63% of patients had a PSA decline of more than 50%, and 87% had any PSA response. Remarkably, five patients (13%) showed a response for over two years. Unfortunately, four patients (10%) dropped out of this study because of (severe) side effects (xerostomia), and five patients (13%) terminated treatment due to lack of response following the first cycle [94]. Sathekge et al. enrolled 73 patients with mCRPC for treatment with 225Ac-PSMA-617 (every eight weeks, most patients received up to two–five cycles). A total of 82% of patients had any PSA response in this cohort, and 70% had a PSA decline of >50%. Grades I and II xerostomia were reported in 85% of the patients, not leading to treatment discontinuation [95]. 225Ac-PSMA-617 could benefit patients who did not respond to prior 177Lu-PSMA-RLT. Several studies included patients previously treated with 177Lu-PSMA-RLT. Yadav et al. prospectively enrolled 28 men with mCRPC to receive 225Ac-PSMA-617 treatment (median of three cycles). A total of 54% of these had prior exposure to 177Lu-PSMA therapy. After the first treatment cycle, 25% of the patients had a PSA decline of ≥50%, which increased to 39% at the end of follow-up. Any PSA decline was found in 78.6%. Patients’ refractory to 177Lu-PSMA less frequently showed a PSA decline of ≥50% than patients with no history of 177Lu-PSMA therapy (26.6% versus 53.8%). Half of the patients reported fatigue and 29% xerostomia (grade I/II) as AE [96]. In the study by Fuerecker et al., 225Ac-PSMA-617 was offered every eight weeks (median of two cycles) to 26 patients with mCRPC who progressed after a median of four cycles of 177Lu-PSMA treatment. In 88% of the patients, any PSA decline was described, and 65% had a PSA decline of ≥50%. Grade I/II xerostomia was observed in all patients, leading to study discontinuation in six patients (23%). The reported hematological AE (grade III/IV) were thrombocytopenia (19%), leucopenia (27%), and anemia (35%) [97]. Although these retrospective studies seem promising, further prospective data is warranted. Unfortunately, the clinical application of 225Ac-PSMA RLT is sparse due to the limited availability of 225Ac [98].In recent years, PSMA PET has gained an increasingly important role in both initial diagnosis and at the biochemical recurrence of disease in patients with prostate cancer. In addition, PSMA PET/CT is being used more frequently during follow-up of the disease to assess treatment response. Aside from targeting the PSMA protein for diagnostic purposes, PSMA may also be a target for combined diagnostics and therapeutic purposes, the Theranostics approach. PSMA radioligand therapy has shown to be an effective and safe therapeutic option for patients with metastatic castration-resistant prostate cancer. Its oncological effect is currently being investigated in patients presenting with metastatic hormone-sensitive prostate cancer.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051169/s1, Table S1: Summary of 177Lu-PSMA-617 studies; Table S2. Summary of 225Ac-PSMA-617 studies.Conceptualization, W.I.L., M.C.F.C., D.E.O.-L., and A.N.V.; Writing—original draft, W.I.L.; Writing—review and editing, M.C.F.C., D.M., D.E.O.-L., A.N.V., R.B., and N.H.H. All authors have read and agreed to the published version of the manuscript.This research was partially financed by Cancer Center Amsterdam, Amsterdam, the Netherlands.Ethical review and approval were waived for this narrative review, due to its nature.The patients whom scans are presented in Figure 1 and Figure 2 gave their written informed consent for data to be used for scientifical purposes.No new data were created or analyzed in this study. Data sharing is not applicable to this article.The authors declare no conflict of interest.A 70-year-old patient with biochemical recurrence after radical prostatectomy (Gleason 3 + 4 = 7, PSA-nadir < 0.1 ng/mL) with a PSA of 0.7 ng/mL at the PET/CT scan time. Restaging 18F-DCFPyL PET/CT detected multiple bone metastases (>10) at low serum PSA value ((A); maximum intensity projection). Transversal 18F-DCFPyL PET (B,E) and fused PET/CT (D,G) images illustrate two bone metastases (os pubis left, red arrow: SUVmax: 9.76; L5 vertebra, blue arrow SUVmax: 8.02) with sclerotic substrate on CT (C,F,H).A 79-year-old patient with CRPC after initial treatment with radiotherapy followed by hormonal therapy. Images illustrate improved detection of bone metastases using 18F-DCFPyL PET/CT compared to bone scintigraphy (4 weeks interval). The PSA level at PET was 23 ng/mL. On bone scintigraphy, faint uptake in the lumbar spine, the right acromioclavicular joint, the sternoclavicular, and hip joints were attributed to degenerative changes (A). Transversal 18F-DCFPyL PET (B) and fused PET/CT (D) revealed two foci (red arrows) with intense PSMA-expression in the right iliac bone (SUVmax: cranial lesion 6.2 and caudal lesion 17) and a sclerotic substrate on CT (C) and were classified as highly suspicious for bone metastases. Maximum intensity projection (E) demonstrated additional lymph node metastases above the diaphragm.Radionuclide properties of Actinium-225 and Lutetium-177. Reference: Sgouros G, Nature reviews (2020); 589–608 [10].Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01170.txt

@@ -0,0 +1 @@
+The immune response to colon cancer (CC) is highly variable among patients and is clinically relevant. In this study, we compared the immune response assessment for early-stage CC, as measured by Immunoscore (IS), to pathologist visual scoring of the CD3+ and CD8+ T-cell densities at the tumor site (T-score). The objectives were to determine the inter-observer agreement between pathologists and the concordance between the two methods. Agreement between pathologists was minimal to weak. Moreover, a weak concordance between the two methods was observed, leading to misclassification of 48% of cases by pathologist scoring. Due to the high level of immune infiltrate heterogeneity resulting in disagreement of interpretation among pathologists, IS is unlikely to be reproduced via non-standardized methods.Adjunction of immune response into the TNM classification system improves the prediction of colon cancer (CC) prognosis. However, immune response measurements have not been used as robust biomarkers of pathology in clinical practice until the introduction of Immunoscore (IS), a standardized assay based on automated artificial intelligence assisted digital pathology. The strong prognostic impact of the immune response, as assessed by IS, has been widely validated and IS can help to refine treatment decision making in early CC. In this study, we compared pathologist visual scoring to IS. Four pathologists evaluated tumor specimens from 50 early-stage CC patients and classified the CD3+ and CD8+ T-cell densities at the tumor site (T-score) into 2 (High/Low) categories. Individual and overall pathologist scoring of immune response (before and after training for immune response assessment) were compared to the reference IS (High/Low). Pathologists’ disagreement with the reference IS was observed in almost half of the cases (48%) and training only slightly improved the accuracy of pathologists’ classification. Agreement among pathologists was minimal with a Kappa of 0.34 and 0.57 before and after training, respectively. The standardized IS assay outperformed expert pathologist assessment in the clinical setting.The important role of immune response to the tumor has been demonstrated in numerous solid cancers [1,2,3,4,5,6,7], including Colon Cancer (CC) [8,9,10,11,12,13,14,15,16], with a high-level of tumor-infiltrating lymphocytes (TILs) being consistently associated with a favorable prognosis. Various methods with different cutoff values have been used to assess immune cell infiltration. Hematoxylin and eosin (H&E) staining of tumor tissue is the most frequently used histochemical stain in clinical and research laboratories. However, with this method, it is difficult to count the number of TILs in cancer cell nests [2]. The reproducibility of the immune response evaluation by visual examination of H&E slides was previously reported and showed a low level of concordance between the 11 expert observers (4% of 268 cases evaluated) [2]. Due to heterogeneity of TILs and the subjectivity of its evaluation on H&E slides by pathologists, such a method was not reliable enough for a therapeutic decision-making process.Therefore, markers with more accuracy and added clinical value are needed. Moreover, consensus recommendations for scoring TILs for diagnostic purposes, translational research, and the clinical trials are required. The integration of the IS assay into pathology clinical practice can help to ensure the higher level of accuracy and efficiency for characterization of immune response [17,18,19,20].We previously showed that of all immune cells involved in the in situ immune reaction, CD3+ and CD8+ T-lymphocyte cells (specific populations of tumor-infiltrating lymphocytes; TILs) provided the optimal combination for prognostic purpose. The accuracy of prediction of survival times for the different patient groups was greater with a combined analysis of the center of tumor (CT) and the invasive margin (IM) regions than with a single-region analysis [21]. CD3 and CD8 were also chosen as markers because of the quality of the staining and the stability of these antigens. We then developed and validated the immune-based international consensus IS assay [2]. Immunoscore® values are reported based on predefined cutoffs and given one of five category scores (IS 0 to IS 4) that are combined into two relevant clinical risk categories: IS Low (IS 0–1) and IS High (IS 2–4). These distinguish tumors with low versus high immune infiltration that are associated with high versus low risk of recurrence, respectively. IS is now recommended for use in conjunction with the TNM classification system to estimate prognosis for early-stage CC patients in the ESMO Clinical Practice Guidelines [22,23]. In a large international study of more than 3500 stage I-III CC patients, in high-risk stage II patients, IS identified a large fraction of patients (70%) whose risk for recurrence was similar to that of low-risk stage II patients when not treated with chemotherapy [3,23,24]. This strongly suggests a clinical utility for the IS assay to identify patients having a low biological recurrence risk despite the presence of pathologic high-risk features that might otherwise trigger adjuvant chemotherapy. These patients may avoid unnecessary treatment and its attendant toxicities. In addition, IS was shown to be a powerful prognostic marker for stage III CC patients in two randomized phase III clinical trials [3,4] and also predicted response of adjuvant chemotherapy in two independent cohorts [4,6].The analytical validation of IS has been demonstrated previously [20,25]. Immunoscore® was deemed to be a robust, reproducible, quantitative, and standardized immune assay, with a high prognostic performance, independent of all of the prognostic markers currently used in clinical practice. The immune response was introduced for the first time into the latest (5th) edition of the WHO Digestive System Tumors as “an essential and desirable diagnostic criteria for CC”. Furthermore, the 2020 ESMO Clinical Practice Guidelines for CC included IS to refine the prognosis, stratify patients according to risk, and thus adjust the chemotherapy decision-making process, although its role in predicting an adjuvant chemotherapy effect is uncertain. Therefore, it is important to compare the performance of the standardized consensus digital pathology IS to an evaluation of the immune response by visual examination of H&E slides or by a visual examination of CD3+- and CD8+-stained slides by expert pathologists.Here, we compared the performance of automated digital pathology (using IS) and pathologist visual scoring of CD3+ and CD8+ T-cell densities at the tumor site (T-score) for assessment of immune response in patients with CC. The performance of each of the two methods in assessing the immune response status and the impact of misclassifications of the risk of recurrence on patient management and treatment decisions was evaluated.This study compared the immune response assessment in early-stage CC by two methods: (1) expert pathologist evaluation of CD3+ and CD8+ stained slides at the tumor site (T-score) in two steps: (i) without training and (ii) with training and (2) artificial intelligence assisted digital pathology (IS).Representative high-resolution scanned images of CD3+ and CD8+ single-stained tumor specimens from 50 patients were selected from the IDEA-France study [4]. The mean densities of CD3+ and CD8+ T-cells quantified in the CT and IM were converted into IS with predefined cutoffs [26,27]. Immunoscore® uses standardized percentile values (0–100%), and the algorithm categorizes the continuous Immunoscore® into five groups (0, 1, 2, 3, and 4). A predefined two-level classification (2 groups of recurrence risk) uses predefined cutoffs corresponding to IS-Low with a mean percentile of 0–25% (IS 0–1) and IS-High with a mean percentile of >25–100% (IS 2–4), consistent with the validated assay cutoffs determined in the Society for Immunotherapy of Cancer (SITC) study [6], with IS-Low indicating a poor prognosis (high-risk of relapse) and IS-High is indicative of a good prognosis.The mean of the 4 percentiles (mP) obtained for CD3+ and CD8+, either in the CT or IM, was calculated for each specimen collected from 50 patients and grouped into 10 categories (0–10%, 10–20%, 20–30%, 30–40%, 40–50%, 50–60%, 60–70%, 70–80%, 80–90%, and 90–100%). Within each category, 5 cases were randomly selected to ensure a uniform distribution of 50 cases along the dynamic range of IS at the level of mP. Then IS was categorized into two groups.The subset of 20 cases, for which IS was around a 25% mP clinical cutoff point, were analyzed separately. This subgroup consisted of 10 cases with IS-Low (≤25%) and 10 cases with IS-High.Four expert pathologists with broad experience in gastrointestinal cancer pathology independently assessed the immune infiltration (CD3+ and CD8+ T-cells) for the 50 selected cases through qualitative visual and manual scoring via an online secured-access web gallery. Pathologists were asked to classify each marker density and to sort them into three categories (Low, Intermediate, and High) and a final 2-class T-score (Low or High) was generated in accordance with clinical reporting. The images were analyzed blindly without training instructions. To avoid a learning bias, cases were analyzed by each pathologist in a pre-specified, individualized, and randomized order.In a second step, the pathologists were trained to assess densities of each marker at the 25% mP (separating Low and Intermediate staining) across a selection of four illustrative images (Table 1). In order to recognize heterogeneity in T-cell infiltrates from different regions in multiple tumors but yielding equivalent T-scores, pathologists were further provided a set of 12 images that represented a spectrum of CD3+ and CD8+ densities across the CT and IM regions (Table 1).The four pathologists repeated immune infiltration evaluation on the same 50 selected cases after this training and reported their classification category for CD3+ and CD8+ T-cells in both the CT and IM and the overall category (Low/Intermediate/High) for each case. A final 2-class T-score (Low or High) was generated in accordance with clinical reporting of the Immunoscore®. The immune infiltration assessment data were captured using a data collection Excel spreadsheet and analyzed.The 50 reference cases were internally analyzed three times to evaluate repeatability of the IS method. The IS module (Immunoscore® Analyzer, Veracyte, Marseille, France) was used for automatic detection of the CT and IM, quantification of CD3+- and CD8+-stained T-cells, and classification of the reference cases into the clinical IS categories. Each IS repetition (identified as DP1, DP2, and DP3) and validation of the results were carried out by two histotechnicians who evaluated the technical parameters, including immunoperoxidase staining quality (the histotechnicians are experienced and expert in performing quality control analysis of IS cases). The IS results and the name of the histotechnician were captured using a data collection Excel spreadsheet.The statistical analysis was used to explore the following types of concordance: between individual pathologist assessment and IS for all cases (n = 50) and for the subset of cases around the clinical 25% IS cutoff (n = 20) before and after training, inter-pathologists’ agreement with visual assessment of T-score, and among three repeated IS assessments.The Cohen’s Kappa coefficient was used to evaluate agreement of Immunoscore® results between the two rating methods, IS and pathologists’ scoring. The Fleiss’s Kappa coefficient test, an extension of the Cohen’s kappa, was used to compute the agreement between multiple observers’ assessments. In accordance with McHugh et al. [28], the level of agreement was categorized according to the Kappa values as none (0–20%), minimal (21–39%), weak (40–59%), moderate (60–79%), strong (80–90%), and almost perfect (>90%). A negative Kappa indicated that there was less agreement than would be expected by chance, given the marginal distributions of ratings.Without previous training, the agreements were weak between pathologists’ T-score classification and the reference IS for the immune infiltration assessment of 50 CC cases (Figure 1, plain dark blue bars). The mean agreement (Cohen’s Kappa) for pathologists’ T-score classification compared to the reference IS was 0.47 (minimum and maximum agreements were (0.29–0.59)). The maximum agreement rate with the reference IS was 82% (Cohen’s Kappa of 0.59) for pathologist #2 and 80% for the three other pathologists #1, #3, and #4, with Cohen’s Kappa from 0.29 to 0.53 (Figure 1, plain dark blue bars). The lowest percentage of negative agreement between T-score and IS, 25%, was observed for pathologist #4, while the lowest positive percent agreement was observed for pathologists #2 and #3 (79%).The disagreement rates for T-score classification versus the reference IS for each pathologist were even higher for the 20 cases with IS percentiles around the clinical cutoff. A minimal level of agreement was reached by pathologists’ visual evaluation compared to the reference IS: the mean agreement (Cohen’s Kappa) for pathologists T-score classification compared to the reference IS was 0.30 (minimum and maximum agreements were (0.10–0.50); Figure 1, plain light blue bars).After training, a moderate level of agreement between the pathologist T-score visual assessment and the reference IS on the 50 cases was reached for one pathologist (#3; Cohen’s Kappa of 0.67) while it remained weak for all other pathologists (Cohen’s Kappa ranging from 0.46 to 0.56). The mean agreement (Cohen’s Kappa) for pathologists’ T-score classification compared to the reference IS was 0.54 (minimum and maximum agreements were (0.46–0.67); Figure 1, dotted dark blue bars).The best agreement rate for classification of the 20 cases around the clinical cutoff after training was observed for pathologist #3 (70%) with a corresponding weak Cohen’s Kappa agreement of 0.40 (versus 0.30 before training; Figure 1, dotted light blue bars).The impact of training was further assessed by evaluating the four different types of “agreement” (i.e., combining concordance or discordance before and after training, Figure 2). On average, training had a positive impact in 20% of the analyzed cases (Type 3). However, training had no impact for 18% of the cases (Type 2) and even worsened the concordance between the visual assessment and IS in 15% of the cases (Type 4).The inter-observer agreement for the 50 selected CC cases into T-score classification was weak before training (Fleiss’s Kappa of 0.34) and was still weak after training (Fleiss Kappa of 0.57; Figure 3). The agreement rates were minimal or nonexistent for the 20 cases around the clinical IS cutoff point (Fleiss Kappa of 0.13 and 0.37, before and after training, respectively; Figure 3).Pathologist disagreement with the reference IS, as defined as the percentage of cases for which at least one pathologist assessment was not concordant with the reference IS, was observed in nearly half of the cases without training (48%; 24 out of 50; Figure 4A) and in 30% of the cases (15 out 50) after training (Figure 4B). The analysis of the 20 CC cases around the IS clinical cutoff resulted in even lower concordance with the overall disagreement rate as high as 80% and 65% for before and after training, respectively. Pathologists agreed only on three High T-scores and one Low T-score out of 20 cases before training (Figure 4A) and on 5 and 2 cases after training (Figure 4B), respectively.The agreement between three repeated IS scores and the initial reference IS score for each of the 50 CC cases is illustrated in Figure 5. Almost perfect agreement was observed (Cohen’s Kappa of 0.93). In the first repeat (DP1, Figure 5), only 1 out of 50 cases were incorrectly classified as compared to the reference IS result and only two more cases were misclassified in the two remaining repeats (DP2 and DP3, Figure 5), leading to an agreement of 94%. The three discordant cases were very close to the cutoff point of 25% with IS mPs ranging from 21.2% to 26.2%. Thus, IS yielded a sensitivity of 95% and a positive predictive value of 97% (overall agreement of 94%).The reproducibility of IS digital pathology was previously assessed [2]. Representative images from five centers (Belgium, Canada, China, France, USA) of tissue stained for CD3+ and CD8+ (n = 36), having IS ranging from lowest to highest (2.5th to 90th percentiles, respectively) were re-analyzed by eight pathologists from different centers. These eight IS digital pathology quantifications revealed a strong reproducibility (mean cell densities in each tumor region, r = 0.97 for tumor; r = 0.97 for invasive margin; p < 0.0001). Only 2.1% variation in the mean percentile of CD3+ and CD8+ T-cell densities was found between IS quantifications. These observations were confirmed in an independent study [20]. This showed the strong reproducibility of IS using digital pathology.Since visual evaluation of tumor infiltrating lymphocytes in H&E-stained slides by pathologists was not sufficiently accurate for clinical decisions and, as it was important to assess the added value of automated digital pathology over visual assessment on the same CD3+ and CD8+ stains, we evaluated the reproducibility of a visual examination on these slides (T-score) by expert pathologists. The inter-pathologists’ reproducibility and the differences between T-score and automated digital IS were evaluated.The IS method was confirmed to be a very robust method that produced reliable and consistent data with a very high degree of agreement (94%) between repeated measures. Moreover, the rare cases of discordance (3 out of 50) were all very close to the cutoff value of 25% and re-testing such samples to correctly assign their score would be simple. In contrast, a significant disagreement was observed for the visual semi-quantitative pathologist T-score (High or Low). This inter-observer disagreement was not improved by providing pathologists with training for the visual scoring process to recognize the IS cutoff points of prognostic importance. Furthermore, the study revealed that the effect of training was heterogeneous between pathologists and that, overall, training only marginally improved and, in fact, for two pathologists, worsened the concordance between the visual assessment and IS. Importantly, a high rate of disagreement was observed when comparing the pathologists’ visual assessment with the reference IS, leading to misclassification of almost half the cases (48%) and this disagreement was particularly high for the cases around the 25% clinical cutoff (80%).The lack of improvement in agreement between pathologist evaluation and quantitative digital pathology, before and after training, is likely multifactorial. In fact, the size of a colon tumor is quite large, and a whole slide analysis revealed a heterogeneous pattern of CD3+ and CD8+ within different areas of the tumor. Furthermore, the mean density of these cells is higher at the invasive margin compared to the core of the tumor, rendering the overall visual evaluation difficult. In addition, these immune cells can be present within the tumor glands or within the stroma at different densities and can be clustered or dispersed even within the same tumor. CD3, encompassing both CD8 and CD4 T-helper cells and CD8 cells, also have different densities in different areas of the tumor, and the evaluation has to be done twice for each of these markers on consecutive slides. Looking at the overall slide is tedious, and the semi-quantitative evaluation of so many heterogeneities is very complex and in fact very subjective. For such evaluation, the novel tool of quantitative digital pathology is clearly much more appropriate, as demonstrated by the poor performance of pathologist scoring, even after training.To illustrate how an incorrect determination of an immune response of stage II and III CC patients could influence the subsequent treatment and potential outcome of the patient, we illustrated a clinical decision tree for these patients (Figure 6) [29]. For patients with stage II CC, the misclassification of patients with IS-Low to highly infiltrated tumors (IS-High) results in patients being identified as stage II CC at low clinical risk when they are in fact at high biological risk. This is important because such a situation would produce false expectations of a low risk of recurrence for these patients who will not be monitored as closely as those at high risk of recurrence to detect signs of relapse earlier. Based on the worst-case negative agreement between visual T-score and IS observed in this study (25%), 75% of IS-Low CC cases would be classified as having a good outcome. Thus, 17% of low-risk stage II or 9% of all stage II patients would not be appropriately considered as high-risk patients. They may be undertreated and under screened. Conversely, misclassification of truly IS-High stage II CC patients as having tumors with low immune cell infiltration could result in patients recommended for adjuvant chemotherapy when their recurrence risk is low, and thus they are unnecessarily exposed to long-term toxicity and side-effects of chemotherapy (Figure 6). In the worst-case scenario observed in this study (positive agreement of 79%), this represents 7% of all stage II CC patients who might be overtreated.In the case of stage III, if IS-Low CC patients were misclassified as IS-High, they would not be identified as poor responders and may be unnecessarily subjected to additional therapy (six months versus three months) and its associated long-term toxicity and intense side effects (Figure 6). Considering the worst incorrect classification observed in this study, 75% of the IS-Low patients would be identified as good responders to extended adjuvant treatment, which represents 37% in the stage III high clinical risk group (T4/N2) or 15% of all stage III CC patients.Finally, in the worst-case scenario observed, up to 21% of IS-High cases might be incorrectly identified as poor responders to six months of chemotherapy (IS Low) and thus be subjected to an increased risk of relapse.Altogether, given an estimated 101,420 and 23,000 new stage II and new stage III CC patients per year, respectively, pathologist visual evaluation of T-score would lead to 8700/5800 (before/after training) CC cases being misclassified annually and possibly receiving inappropriate patient care.A limitation of the study relates to the sample size and to the relatively low number of expert pathologists who evaluated the CD3+- and CD8+-stained images. These results should be validated with a larger cohort of patients and with a larger number of expert pathologists. However, given the very important difference between pathologist T-score classification and the reproducible IS quantification, these results confirm the importance of new tools for pathologists, namely quantitative digital pathology.The potential negative impact that misclassification of immune response assessment and thus erroneous prognosis and risk evaluation might have on the clinical management of patients with CC was shown to be significant.Our results showed that the IS assay provided the best stratification of patients into prognostic recurrence groups (low versus high). We conclude that the standardized and robust IS assay outperforms the assessment of expert pathologists in the clinical setting for immune response evaluation and can thus provide the most appropriate individualized therapeutic decisions for patients with CC.Conceptualization: I.B., F.H., J.G. and A.C.; methodology: I.B., A.K., A.L., J.G., F.H. and A.C.; digital pathology and visual assessment realization: A.L., M.L., Y.L., A.T. and F.A.; writing: I.B., A.K., F.H., J.G. and A.C.; statistical analysis: A.K. All authors contributed to manuscript revision as well as read and approved the submitted version. All authors have read and agreed to the published version of the manuscript.This work was supported by Veracyte.The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee of Groupe Hospitalier Pitié Salptrière (EudraCT: 209-010384-16 approved on 10 April 2009).Informed consent was obtained from all subjects involved in the study.Authors agree to make data and materials supporting the results or analyses presented in their paper available upon reasonable request.The authors would like to acknowledge and thank Alexandre Chollat-Namy, Nicolas Brandone, and Sergiu Coslet for histology services. The authors thank Magdalena Benetkiewicz and Graham Poage for editing assistance. We thank the GERCOR who allowed the use of CD3 and CD8-stained images and Immunoscore® analytical results from the IDEA-France ancillary biomarker analysis (ClinicalTrials.gov ID NCT03422601). Immunoscore® is a registered trademark of Inserm.The authors declare no conflict of interest.Bar plot showing agreement between individual pathologist visual assessment (T-score) and the reference Immunoscore® (IS) before (plain bars) and after (dotted bars) training. The y-axis shows the level of agreement according to the Cohen’s kappa value: none (0–0.20), minimal (0.21–0.39), weak (0.40–0.59), moderate (0.60–0.79), strong (0.80–0.90), and almost perfect (0.91–1). For each pathologist (Patho 1–4), T-score was expressed as High or Low for 50 cases along the dynamic range of IS (dark blue) or 20 cases around the clinical cut-off of 25% (light blue).Figure 2. Distribution of agreement types between T-score visual assessment and the reference IS (Low, Intermediate, High) in a set of 50 colon cancer cases, before and after training (average of cases falling in each type is reported in parentheses).Fleiss’s Kappa values for inter-observer agreement for T-score classification (50 selected CC cases and 20 cases around the clinical IS cutoff point) before and after training.Graphical plot representing the agreement between each of the four pathologists visual T-score and IS before (A) and after training (B) of 50 colon cancer cases. Reference IS scores (Low and High) from 50 colon cancer cases (x axis) are plotted against the pathologist visual T-score and IS methods (y axis). Dark green circles indicate an agreement between all pathologists and the reference digital pathology IS method. Bright orange circles indicate disagreement between at least one of the pathologists and the reference IS. The mean percentiles (mP) of the CD3+ and CD8+ T-cells densities are represented as circles, whose size is proportional to the mP value observed for each case. The 50 cases were ranged from the lowest mP to the highest mP and IS was translated into 2-category classification (dashed line): Low IS (mP ≤ 25%) and High IS (mP ˃ 25%); the T-score classification for each pathologist is represented by blue circles with a black outline (Low T-scores) and yellow circles with a black outline (High T-scores). Abbreviations: Patho, pathologist; mP, mean percentile; DP, digital pathology.Graphical plot representing the agreement between three repeated IS analyses and reference IS assessment for the 2-category classification. The IS scores for 50 colon cancer cases (x axis) are plotted against the scoring method (horizontal lines, from the bottom to the top, represent three repeated IS analyses [DP1_IS, DP2_IS, and DP3_IS], the reference IS) used (y axis). The mean percentiles (mP) of the CD3+ and CD8+ T-cell densities are represented as circles, whose size is proportional to the value observed for each case. The cases range from the lowest mP to the highest mP and are translated into IS with the 2-category classification: Low IS (mP ≤ 25%) and High IS (mP ˃ 25%). Abbreviations: mP, mean percentile; IS, Immunoscore®; DP_IS, digital pathology Immunoscore®.Decision tree for patients with stage II (A) and III (B) colon cancer considering the IS-High and Low scoring. Abbreviations: 6m, 6 months; 3m, 3 months.Pathologist IS training session steps.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01171.txt

@@ -0,0 +1 @@
+Chemotherapy and chemoradiotherapy are only effective in 25% to 30% of patients with oesophageal cancer. Being able to predict which patients will respond to chemotherapy and chemoradiotherapy before they receive this treatment will prevent patients undergoing unnecessary procedures that may reduce their quality of life and help source alternative treatment options faster. The scope of this review was to understand whether microRNAs, small non-coding RNA molecules that regulate gene expression, can be used as biomarkers to predict a patient’s response to chemotherapy and/or chemoradiotherapy treatment. This review showed that a number of microRNAs may have the potential to predict response to chemotherapy and chemoradiotherapy alongside other pre-treatment features already used. More research is needed to translate the use of microRNAs as biomarkers of response to the clinical setting, as well as understanding the effects different types of treatment have on predictability.Oesophageal cancer (OC) is the ninth most common cancer worldwide. Patients receive neoadjuvant therapy (NAT) as standard of care, but less than 20% of patients with oesophageal adenocarcinoma (OAC) or a third of oesophageal squamous cell carcinoma (OSCC) patients, obtain a clinically meaningful response. Developing a method of determining a patient’s response to NAT before treatment will allow rational treatment decisions to be made, thus improving patient outcome and quality of life. (1) Background: To determine the use and accuracy of microRNAs as biomarkers of response to NAT in patients with OAC or OSCC. (2) Methods: MEDLINE, EMBASE, Web of Science and the Cochrane library were searched to identify studies investigating microRNAs in treatment naïve biopsies to predict response to NAT in OC patients. (3) Results: A panel of 20 microRNAs were identified as predictors of good or poor response to NAT, from 15 studies. Specifically, miR-99b, miR-451 and miR-505 showed the strongest ability to predict response in OAC patients along with miR-193b in OSCC patients. (4) Conclusions: MicroRNAs are valuable biomarkers of response to NAT in OC. Research is needed to understand the effects different types of chemotherapy and chemoradiotherapy have on the predictive value of microRNAs; studies also require greater standardization in how response is defined.Oesophageal cancer (OC) is the ninth most common cancer diagnosed globally, yet the sixth most common cause of cancer related death [1], resulting in an estimated 436,000 patient deaths in 2017 [2,3]. Five-year survival for OC is 17% [4]. The prognosis is bleak when compared to other cancers such as colorectal cancer, of which the 5-year survival is almost 3 times greater than that of OC [5]. These poor outcomes can at least in part be attributed to late presentation, with 47.9% of patients diagnosed at stage 4 [6]. OC is more common in men who are 3 to 4 times more likely than women to develop OC, and the median age of diagnosis is 68 years [1]. There are two major OC subtypes, oesophageal squamous cell carcinoma (OSCC) and oesophageal adenocarcinoma (OAC), for which OSCC estimated to contribute for over 70% of all OC diagnoses globally [7,8]. OSCC, shows greater prevalence across Asia within an “oesophageal cancer belt” that stretches from north-east Iran to north-west China [9]. This is likely due to the higher prevalence of tobacco use, along with genetic differences in alcohol metabolism which leads to acetaldehyde accumulation, a known carcinogen [7,10]. OAC is, however, predominant in the Western world, probably due to the high incidence of obesity and gastro-oesophageal reflux disease [11] which are the primary risk factors for OAC development. The UK has the highest incidence of OC in Europe with approximately 10,000 new cases per year [12]. The UK’s 5-year survival for OC is 12% in line with average across Europe; however this varies largely from 9% in Latvia to 25% in Belgium [4].OAC is most often found in the lower part of the oesophagus and at the gastro-oesophageal junction, where it frequently develops from its precursor Barrett’s oesophagus [13]. Persistent exposure to acid and bile reflux, results in mucus-secreting glandular metaplasia [14]. Increased genetic mutations and loss of heterozygosity are seen during epithelial proliferation. Notably in the progression to OAC two defining genetic mutations are present that can be utilised to differentiate between non-dysplastic Barrett’s oesophagus, high grade dysplasia and OAC tissues, these are TP53 and SMAD4. Mutations in the tumour suppressor gene TP53 occur in 50–71% of all OAC cases [7] but have been shown to be recurrently mutated in high grade dysplasia (HGD) and OAC samples. SMAD4 mutation is shown to be exclusive to OAC tissue and thus it can be concluded it lies at the boundary between progression from HGD to OAC [15]. The loss of these tumour suppressor genes (TSG) or mutation of proto-oncogenes leads to dysplastic Barrett’s due to a greatly increased rate of uncontrolled cellular proliferation [7]. Further mutagenic changes and chromosomal instability over time result in the formation of OAC and without intervention, infiltration of the basement membrane and subsequent metastasis.OSCC pathology is subtly different; it more commonly presents in the middle and upper parts of the oesophagus [16]. Persistent physical insults, primarily chronic alcohol and tobacco use, lead to squamous cell hyperplasia [17]. Over time should exposure be unchanged, genetic mutations will accumulate due to greatly increased rates of cellular proliferation. TP53 mutation in OSCC is almost universally found in all patients, approximately 92% [7,18]. However, NOTCH1 and NOTCH3 mutations have been shown to be significantly more frequent in OSCC than OAC, at 33% versus 25%, respectively [18]. These mutagenic accumulations eventually result in a loss of negative feedback mechanism due to malfunction of tumour suppressors. The result is uncontrolled rapid rates of cellular proliferation producing a carcinoma [7].Currently there is no national screening programme for OC in the UK [19], in contrast to Asian countries such as Japan and Korea where a higher disease incidence is seen. These have been shown to significantly improve outcomes [20,21], likely due to earlier detection and thus earlier intervention. Screening programmes consist of fibreoptic upper gastrointestinal endoscopy or Barium Upper Gastrointestinal Series for those over 40 years, recommended every 2 years thereafter [20,21]. With these interventions approximately two thirds of upper GI cancers are detected at an early stage, compared to the 70% of late-stage diagnoses across Europe [1,20,21]. The largest issue with screening for OC is the invasiveness of this procedure and therefore the risk it poses to the patients [19], particularly should these results be negative and therefore futile. Mass screening would therefore place 1 in 200 to 1 in 10,000 patients at risk of adverse events such as infection, perforation and bleeding [22]. The American and British Societies of Gastroenterology suggest screening for OAC should be provided to patients with reflux > 5 years, Caucasian males and family history of Barrett’s oesophagus or OAC [7]. Despite recent evidence showing one off endoscopic screening in China led to a reduction in incidence and mortality, no current OSCC screening is in place [23] possibly due to its lower population prevalence in the UK. Traditionally, there have been no minimally invasive procedures with high enough sensitivities to consider their use in widespread screening programmes. However, new research into Cytosponge™ technology for detection of Barrett’s oesophagus and early dysplasia suggests a specificity and sensitivity of 79.9% and 92.4% respectively, which is comparable to current screening programmes for colorectal cancer in the UK with a false positive rate of 2–9% [24], as well as improved detection rates of Barrett’s oesophagus in the primary care setting [25]. Utilisation of this minimally invasive sampling method, alongside modern genetic testing could prove to be a highly sensitive and specific way of detecting and tailoring treatment regimens to patients.Where OC is potentially curable at presentation, it is locally advanced in the majority of cases. Standard of care treatments in this setting for both OSCC and OAC are usually extensive and invasive, requiring neoadjuvant chemotherapy or chemo-radiotherapy (NAT) followed by surgical resection as recommended by NICE guidelines [26]. Guidance on the treatment pathway of OSCC and OAC based upon staging and functional assessment of the patient, as recommended by the European Society of Medical Oncology are outlined in Figure 1 [27]. A variety of studies have investigated the benefits of NAT. The MRC MAGIC trial showed patients who receive a neoadjuvant regimen of epirubicin, cisplatin and 5-fluorouracil (5-FU) (ECF) therapy had a higher rate of overall survival (5-year survival, 36% vs. 23%) and progression free survival (0.53 to 0.81, p < 0.001) in comparison to patients undergoing surgery alone [28]. Similarly, the CROSS trial demonstrated that neoadjuvant chemoradiotherapy improved median overall survival from 24 to 49.4 months vs. surgery alone [29]. However, only 25% to 30% of patients achieve a partial or complete pathological response [30,31], and it carries a 0.5 to 2% mortality rate [32]. Early identification of patients that respond well could improve outcomes by preventing the administration of treatment regimens that are unlikely to be effective and facilitating treatment modulation [33]. Response to therapy is usually assessed via assignment of the Mandard Tumour Regression Grade (TRG) ranging 1 to 5 [34]. Responders are usually defined as TRG 1 (complete regression with no viable tumour cells evident) and TRG 2 (presence of residual cancer cells), at least for patients receiving chemotherapy. By administering NAT in patients who do not respond well, surgery is delayed, which if carried out earlier may have proven more effective. The main benefits of NAT are the increased chance of complete resectability of the primary tumour, as reduced tumour mass induced by NAT decreases the area of resection required, as well as improved prognostic outcome due to the decreased incidence of nodal micrometasteses [32,35,36]. On the contrary, tumour progression during therapy can occur in those patients who do not respond well to NAT or conversely overtreatment of tumours with a favourable prognosis that are unlikely to respond to NAT. Therefore, identifying biomarkers that allow successful identification of who will or will not respond to therapy are desperately needed to allow rational treatment decisions to be made.MicroRNAs (miRNAs) are single stranded noncoding RNA molecules approximately 22 nucleotides long, regulating gene expression at the transcriptional or post-transcriptional level [36,37]. They do this by binding in a specific sequence to the complementary region in the 3′ untranslated mRNA region, which then regulates the translation of mRNAs to proteins [38]. miRNAs can bind to a complementary mRNA region resulting either in blockage of translation or degradation of a section of miRNAs via the RNA-induced Silencing Complex (RISC) complex, both leading to inactivation of a gene (Figure 2) [37,39]. Common molecules regulated are signalling proteins as well as transcription factors to RNA binding proteins [40]. miRNAs play an important role in biological pathways and their expression is dysregulated in multiple pathological mechanisms [41]. Aberrant miRNAs expression patterns are involved in the initiation and progression of oncogenesis due to their role as TSG and oncogenes. Oncogenic miRNAs target and prevent the expression of endogenous TSG, which activate pathways associated with OC, such as the reduced expression of miR-27a leading to permanent activation of the KRAS pathway [42]. TSG are often downregulated, dysfunctional or completely lost in OC such as miR-30b-5p [43] or miR-34a [44] in OSCC whereas those associated with proto-oncogenes are upregulated [45].Recent research into the effects of NAT on miRNAs expression in other cancers such as breast cancer and rectal cancer have shown promising results. Over the course of NAT Lindholm et al. (2019) [46] showed that tumour suppressor miRNAs expression, such as miR-100-5p and miR-125b, were upregulated following treatment [46] which may reflect a role in the regulation of chemosensitivity. Kheirelseid et al. (2013) [47] identified expression signatures of miR-16, miR-590-3p and miR-561 that were predictive of complete versus incomplete response to neoadjuvant chemoradiotherapy in pre-treatment samples of rectal cancer [47]. A recent study showed that several miRNAs can predict poorer overall survival in both OSCC and OAC [48] such as the upregulation of miR-21 and downregulation of miR-133a. Hence, using miRNAs as predictors of pre-treatment response as well as other factors such as survival, seems to be a viable non-invasive potential solution to improving the accuracy of patient allocation to treatment.Currently, there is no biomarker for stratifying patients into responders and non-responders to NAT using treatment naïve samples. Utilising circulating miRNAs to predict response to NAT could provide a simple and minimally invasive solution such as blood sampling could be used to identify patients that will benefit from NAT, thus improving outcome and quality of life. This could be utilized to facilitate optimal treatment choices and the likelihood of patients achieving a pathological good or complete response. Here we review 15 studies that investigated whether miRNAs could be used as a novel biomarker for predicting the response to neoadjuvant therapy, via the use of pre-treatment sample analysis in patients with OAC or OSCC.Online databases including MEDLINE, EMBASE, Web of Science and Cochrane library were searched to identify relevant studies investigating the use of miRNAs to predict response to NAT in OAC and OSCC (Figure 3). Articles were selected based on the following criteria. All participants must be ≥18 years diagnosed with locally advanced OAC or OSCC; treatment via neoadjuvant chemotherapy or chemoradiotherapy followed by surgical resection; histological samples were pre-treatment samples, the rationale being that assessing post-treatment samples would not “predict” response but simply evaluate a response that has already occurred. Standard dataset restrictions were placed upon each database, these being English language only and from 1 January 1990 onward. The search strategy devised was modified as per each database syntax requirements with MeSH terms being utilised for MEDLINE and Cochrane library databases. The main concepts included in this search were oesophagus/gastro-oesophageal junction, cancer, response, chemotherapy/chemoradiotherapy and miRNA. Boolean AND/OR operators along with truncation and wildcard syntax were then used to link and expand search terms.Out of the 15 studies identified five studies looked exclusively at patients with OAC histology. Three studies utilised frozen OAC samples [49,50,51], one used Formalin-Fixed, Paraffin-Embedded (FFPE) samples [52] and the last study did not define the environment histological samples were stored under [45]. Four articles studied patients treated with neoadjuvant chemotherapy only and one studied neoadjuvant chemoradiotherapy patients. Table 1 describes the key results from the 5 identified studies in OAC, which will be described in detail in this section.Bibby et al. (2015) utilised pre-treatment OAC biopsy specimens (n = 18) to screen for miRNAs by miRNA profiling arrays [49]. MiRNA expression was then assessed via quantitative polymerase chain reaction (qPCR). Patients were administered Cisplatin and 5-Fluorouracil followed by 1.87 Gy/min radiation prior to surgery. Out of 742 miRNAs expressed intratumorally, 67 were differentially expressed between responders and non-responders. MiR-330-5p expression was most differentially expressed between responders (n = 9) and non-responders (n = 10), showing significantly elevated expression levels in pre-treatment samples of responders (p < 0.01). Utilisation of clonogenic survival assay to assess alterations of cell line (OE19 and OE33) sensitivity to chemoradiotherapy showed a statistically significant increase in radio-resistance of cell lines with miR-330-5p silencing. Thus, it concluded that miR-330-5p downregulation may act as a potential biomarker for predicting complete pathological response in patients receiving neoadjuvant chemoradiotherapy.Lynam-Lennon et al. (2016) analysed pre-treatment frozen biopsies from 18 patients receiving Cisplatin and 5-Fluorouracil followed by 40.05 Gy total radiation dose in 15 daily fractions [50]. A TaqMan miRNA assay was used to analyse the abundance of 742 miRNAs, of which 67 were differentially expressed between responders (n = 8) and non-responders (n = 10). Of these samples, expression of 35 miRNAs were increased in non-responders, classified as TRG4 and TRG5. Of particular note is miR-187, which was expressed in every tumour sample; it was significantly lower (p = 0.005) in patients classified as non-responders when compared with responders. Another similar study by Lynam-Lennon et al. (2012) in frozen OAC samples (n = 19) showed miR-31 to be significantly higher in ‘good’ responders (n = 9) defined as patients with TRG 1-2, when compared to ‘poor’ responders (n = 10) (TRG 4–5) [51].Finally, Skinner et al. (2014) carried out a 3-step study across 3 patient cohorts (total n = 118) treated with Cistplatin and 5-Fluorouracil [45]. The discovery cohort (n = 10) were used to identify miRNAs present in resected specimens. A total of 754 miRNAs were examined in pre-treatment biopsies of 10 patients in the discovery cohort via TaqMan array. Median expression of each of these miRNAs in tumour samples was determined and compared between complete responders (n = 5) and non-complete responders (n = 5). Forty-four miRNAs most able to discriminate between these groups entered the model cohort (n = 43) for further analysis via a Fluidigm array. Of these 44 miRNAs, 4 (were miR-505; miR-99b; miR-451 and miR-145) were found to be significantly differentially expressed (p = 0.008) between pathological complete response (pCR) and non-pCR, showing elevated expression levels in non-responders. A logistic regression classifier combining these 4 miRNAs (termed the miRNAs Expression Profile (MEP)) and clinical variables was derived and then validated on a separate cohort of 65 patients. This classifier achieved good accuracy in discriminating between patients achieving a pCR and those who did not, with an area under the receiver operator characteristic curve (AUROC) of 0.77 seen.Ilhan-Mutlu et al. (2015) assessed response to neoadjuvant chemotherapy (nChemo) regimen only with Docetaxel (n = 8); Mitomycin/Cisplatin/5-FU (n = 8); Cisplatin/5-FU (n = 10); Epirubicin/Oxaliplatin/Capecitabine (n = 1) and Others non-specified (n = 6), in 36 OAC patients by quantifying miR-21 and miR-148a levels in pre-treatment surgical specimens [52]. Pathological complete response of patients was later defined using the Mandard Tumour Regression Grade, TRG 1 (complete regression), and no significant difference in expression levels of miR-21 and miR-148a were noted between responders (n = 1) and non-responders (n = 35). Thus, they concluded miR-21 and miR-148a were not predictors of response to NAT in OAC. However, due to the small sample of only one responder to compare the miRNAs no conclusions can be drawn.Of the 15 studies identified utilising OSCC samples exclusively (Table 2), one study used frozen and FFPE OSCC tissue samples [53] while the rest investigated circulating miRNAs in OSCC patients [54,55,56,57,58,59]. Five articles studied patients exclusively treated with nCRT and the other three studied patients treated with nChemo only.Han et al. (2019) examined pre-treatment serum samples of 104 OSCC patients and found that high miR-338-5p expression predicted pathological complete response. Low histopathological response was classified as grade 1a (more than two thirds residual tumour cells) or grade 0 (no significant response to chemotherapy) according to the guidelines of the Japan Esophageal Society (JES) [55,56,60]. Pre-treatment serum miR-330-5p concentrations were inversely correlated with post-therapy pathologic ypT-stage (p = 0.034), ypM (p = 0.014) and overall pathological ypTNM stage (p = 0.017). Chan et al. (2017) [54], Niwa et al. (2019) [58] and Wen et al. (2016) [53] produced results related to variants of miR-193, these were miR-193b, miR-193b-5p and miR-193b-3p. There structures are identical yet originate from different arms of the same pre-miRNA, with one usually being in higher cellular abundance than the others [61]. Chan et al. (2017) prepared pre-treatment serum specimens from blood samples of OSCC patients (n = 47) receiving Cisplatin and 5-Fluorouraccil followed by 40 Gy at 2 Gy per fraction radiation dosage [54]. Analysis showed high pre-treatment serum miR-193b levels predicted pCR (p < 0.001), defined as patients with 0% viable tumour cells at the post-treatment stage. AUROC analysis showed strong predictive power in differentiating between responders (n = 24) and non-responders (n = 23) (AUROC = 0.89, 95% CI: 0.79–1.00, p < 0.001), although this was not assessed on any external data. Findings from a study by Wen et al. (2016) comprised of 106 OSCC patients [53], supported observations by Chan et al. (2017), showing that miR-193-3p was downregulated in non-responders (n = 17) versus responders (n = 89), whereas miR-152, 145-5p and 376a-3p were upregulated in non-responders. An SVM–RBF (Support Vector Machine with Radial Kernel Function) model showed excellent discrimination in distinguishing responders from non-responders using pre-treatment expression of serum miRNAs, with an AUROC of 0.86 (95% CI 0.77–0.97).Conversely to the results of both Chan et al. (2017) [54] and Wen et al. (2016) [53], Niwa et al. (2019) [58] suggested that pre-treatment serum expression of miR-193b-5p (p = 0.004) and miR-873-3p (p = 0.001) were both significantly higher in non-responders (n = 69) compared to responders (n = 15). AUROC values for Niwa et al. (2019) Logistic regression models were here less discriminative with AUROCs of 0.61 (95% CI 0.47–0.74) for miR-193b-5p and 0.68 (95% CI 0.54–0.81) for miR-873-3p respectively, representing fair performance only and no validation on cases not used to derive the models [58].Komatsu et al. (2016) [60] and Kurashige et al. (2012) [57] both produced results related to miR-21 that supported its predictive value. Komatsu et al. (2016) [60] took pre-treatment plasma and tissue samples (n = 37) from patients undergoing Cisplatin and 5-fluorouracil chemotherapy only. Low histopathological response was classified as JES grade 0 or 1a. MiR-21 expression was found to be significantly higher (p = 0.0416) in patients with poor histopathological response (n = 17), vs. responders (n = 13), with fair ability to discriminate cases (AUROC 0.68). Kurashige et al. (2012) showed that in 71 OSCC patients a significant decrease in miR-21 expression during therapy indicates a higher likelihood of pathological complete response or partial response (p = 0.003) [57]. Response was defined according to RECIST 1.0 criteria where pathological complete response is defined as complete disappearance of all tumour lesions on post NAT (pre-surgery) imaging. Thus, the results of both studies were in concordance and suggestive that low pre-treatment concentrations of miR-21 are predictive of a good response to NAT [57,60].Tanaka et al. (2013) studied pre-treatment serum samples from OSCC patients (n = 64) receiving either Adramycin, Cisplatin and 5-Fluorouracil (ACF) or Docetaxel, Cisplatin and 5-Fluorouracil (DCF) [59]. Serum concentrations of miR-200c was only found to be significantly lower in responders (n = 39) compared to non-responders (n = 25) for the ACF treatment group (p = 0.012) versus DCF (p = 0.7167). Complete response was defined as total regression of the tumour.Finally, Komatsu et al. (2016) found pre-treatment plasma concentration of miR-23a were significantly higher in patients with a low histopathological response (n = 24, p = 0.0345) versus a high histopathological response (n = 13 p = 0.0125) [56]. AUROC analysis showed relatively strong differentiation between high and low histopathological response groups with an AUROC of 0.70. Logistic regression analysis also revealed that high pre-treatment miR-23a concentrations proved to be an independent risk factor for chemoresistance (p = 0.0213; OR: 12.4; 95% CI 1.45–105.8).Two studies looked at both OAC and OSCC histological subtypes (Table 3), both utilised FFPE samples. One articled studied patients treated with nChemo only while the other studied patients treated with nCRT.Ko et al. (2012) (n = 25; 80% OAC, 20% OSCC) took pre-treatment specimens from patients receiving Cisplatin and Irinotecan followed by 50.4 Gy total dose radiotherapy [62]. Levels of miR-296 were 2.5 times lower (p = 0.007) and miR-141 was 2 times higher (p = 0.019) in pre-treatment specimens of patients achieving pathological complete response (n = 8). Complete response was defined as 0% viable tumour cells remaining. No logistic regression or AUROC was performed to test accuracy of prediction.Odenthal et al. (2013) [63] (n = 80; 52.5% OAC, 47.5% OSCC) used pre-treatment biopsies and showed pre-therapeutic intratumoral expression of miR-192 and miR-194 were significantly higher in major responders (n = 11) with OSCC only (p = 0.005, p = 0.001), classified as Cologne Regression Grade (CRG) III (near complete regression with <10% VRTC) or Grade IV (complete regression/pathologic complete response) [63].There is minimal research in the areas of in vitro functional genomics and animal models pertaining specifically to the use of miRNAs in the prediction of oesophageal cancer treatment. However, the few studies that are available have shown promising results which often reproduce results seen in vivo from previously discussed studies in this review (see Table 1, Table 2 and Table 3), particularly relating to miR-193b, miR-21 and miR-200c.Hummel et al. (2014) investigated miRNA expression in 5-Fluorouracil and Cisplatin chemotherapy resistant cell lines for OAC and OSCC compared to their chemotherapy sensitive variants (OE19 and KYSE410) [64]. MiR-193b-3p was found to be most significantly upregulated in 5-Fluorouracil resistant OAC cell lines with a 1.54-fold increase when compared to controls (p ≤ 0.05). These changes were consistent with negative post-transcriptional control of KRAS gene expression with its associated mRNAs having a 1.24-fold reduction in expression (p = 0.036). The activity of this miRNA in post-transcriptional control with a direct impact on chemosensitivity of the cell, suggests miRNAs do play a mediatory role in chemotherapy resistance. These results support the human data produced by Niwa et al. (2019) [58] and conflict the evidence produced by Chan et al. (2017) [54] and Wen et al. (2016) [53]. However, both Hummel et al. (2014) [64] and Niwa et al. (2019) [58] utilised nChemo only, whereas both Chan et al. (2017) [54] and Wen et al. (2016) [53] utilised nCRT. Thus, the upregulation of miR-21 in neoadjuvant chemotherapy is associated with non-response whereas its upregulation in neoadjuvant chemoradiotherapy was associated with complete pathological response [53,54,58,64].Hiyoshi et al. (2009) took 20 matched normal oesophageal epithelial samples and oesophageal OSCC samples, as well as seven OSCC cell lines (TE6, TE8, TE10, TE11, TE12, TE14 and KYSE30) to evaluate the role of miR-21 and the effect of anti-miR-21 transfected cell lines [65]. Results showed 18 of the 20 matched OSCC samples overexpressed miR-21 versus normal epithelium (p < 0.001). All seven of the cell lines transfected with anti-miR-21 showed a significant reduction in cellular proliferation and invasion, measured via TaqMan real-time PCR of cell lines T6 (1.8-fold reduction), TE8 (1.25-fold reduction) and TE10 (5-fold reduction).Mahawongkajit et al. (2020) examined two sets of cultured 5-FU resistant OSCC cell lines, TE10-5-FUR and TE11-5-FUR [66]. Each cell line was then compared to its control parent cell line (5-FU sensitive) via a miRNA microarray to determine differences in miRNA expression profiles. Results showed sets of miRNAs displaying the same responses in both cell lines. MiR-146a and miR-483-5p were both significantly upregulated in TE10-5-FUR and TE11-5-FUR cell lines with a 5.59- and 6.19-fold increase, respectively, versus parent control cell lines. For the same cell lines, most significantly in contradiction with Tanaka et al. (2013) [59], miR-200c was collectively downregulated in both 5-FU resistant cell lines by a factor of 4.10. This may be explained by the different types of chemotherapy agents used between the two studies. Mahawongkajit et al. (2020) used cell lines resistant to 5-FU only and thus 5-FU as its only chemotherapeutic agent [66], in contrast to Tanaka et al. (2013) [59] who administered a combination therapy of either ACF or DCF [59]. 5-FU is a potent inhibitor of thymidylate synthetase (TS), which is a key player during thymidylate biosynthesis and hence an essential precursor for DNA synthesis [67]. miR-200c has also be shown to cause inhibition of the same enzyme [68]. Therefore, when 5-FU is used in isolation downregulation of miR-200c would be expected due to the lack of active TS to be inhibited. Thus, it could be concluded that downregulation of miR-200c predicts complete response in neoadjuvant combination chemotherapy via drugs such as Adriamycin, cisplatin and docetaxel used in tandem with 5-FU, as originally suggested by Tanaka et al. (2013) [59].The ability to predict OC in patients that will respond to NAT is vital to improve clinical outcomes whilst reducing treatment associated morbidity. This review demonstrates the potential utility of miRNAs as biomarkers of response (Figure 4), with miR-193b, miR-21 and miR-200c showing the most promising results. However, the utilization of singular miRNAs to predict response to NAT is unlikely to be as sensitive or specific as looking at the miRNA expression profiles of multiple miRNAs. The findings of Wen et al. (2016) [53] and the utilization of machine learning models such as SVM–RBF are likely to prove most beneficial, with the latest research showing excellent ability for machine learning techniques to predict events such as recurrence of OC after surgery [69] when looking at postoperative histopathological characteristics. This same model could be utilized preoperatively, looking at miRNA expression profiles to discern whether patients are likely to respond to NAT.In clinical practice, there are currently no means of stratifying patients into responders and non-responders to NAT in OC. The implementation of miRNA screening prior to the initiation of NAT would allow for patients and the healthcare professionals supporting them to make a more informed decision as to whether treatment is likely to prove beneficial. This is important due to the risks of NAT and the effect it has on the limited length of time and quality of life patients may have. By utilizing a method of predicting response to NAT, such as miRNA screening, in conjunction with new non-invasive diagnostics such as Cytosponge™ technology [25,70] a minimally invasive widespread screening programmed in those at high risk of OC could be formulated. However, research would be needed to understand the ability to predict response utilizing miRNAs from pretherapeutic Cytosponge™ samples.Despite the number of biomarkers discovered and studied, less than 0.1% are utilized in clinical practice [71]. Most often, this is a result of restricted study design and insufficient sample size or representation. In practice, utilization of miRNAs could surpass these clinical hurdles via the use of multiple different models for miRNA expression in patients with OAC versus OSCC and those receiving nChemo versus nCRT. Establishing the clear differences in miRNA expression between treatment types and doses, and linking these with OC histology, is a key step in establishing miRNAs as clinically viable biomarkers. For each set of treatments, a well-designed study with a large sample size and accurate measurements of predictability, such as AUROC, would prove robust enough for potential implementation into practice. Despite this, there are considerable obstacles for the application of pre-treatment miRNA testing within clinical practice. For example, standardisation in the extraction of miRNAs from either tissues or body fluids. Studies have shown miRNA concentrations in samples can not only differ between tissues and bodily fluids but also be directly affected by the method of extraction itself [72,73]. Further to this, when assessing circulating fluid samples miRNA concentration differs between subfractions (e.g., whole blood, peripheral blood mononuclear cells and plasma), thus it is important to standardise the method of extraction and the subfraction from which miRNAs are to be studied [74]. Studies identifying circulating miRNAs that can predict response to NAT are much more likely to be valuable in clinical practice due their greater accessibility and obtaining these samples is less invasive compared to tissue biopsies. Articles reviewing intratumoral miRNA concentrations are more useful in determining the functional mechanisms by which miRNA expression links to how patients respond to nChemo/nCRT. In addition to this, assessing miRNA expression in vitro often leads to differential results between samples due to the interplay between miRNA expression and the intratumoural microenvironment [49,75]. Based on this, future translational research must focus on the standardisation of miRNA sampling and extraction in circulating fluids, in order to become robust enough biomarkers to use in clinical practice.This review has provided evidence suggesting that miRNAs may be robust biomarkers for predicting response to NAT by differentiating between responders and non-responders. These studies suggested miR-505, miR-99b, miR-45 (for OAC patients) and miR-193b (for OSCC patients) are accurate biomarkers for predicting response to NAT. In the literature presented in this review the three miRNAs that have consistently appeared significant in both histological OC subtypes, namely OAC and OSCC, are miR-21, miR-193b and miR-200c thus their overall function in OAC and OSCC and other common cancers is discussed.MiR-21 is a commonly dysregulated in a wide variety of cancers such as renal carcinoma, non-small cell lung cancer, gastric cancer, colon cancer and breast cancer [76]. In oesophageal cancer, high miR-21 has been associated with increased stromal fibroblast activity and increased cell migration [77]; therefore, it is thought to act as an oncogene during the neoplastic life cycle of OSCC with its function being less clear in OAC [76,78]. Studies suggest miR-21 is a useful biomarker in the prediction of response to other cancers such as HER2 positive breast cancer and colorectal cancers [79,80].The absolute role of miR-193b in oesophageal cancer is not fully understood, despite the miRNA being known to act as a TSG in various types of gastric and colon cancer [81,82]. Various studies have shown miR-193b initiates apoptosis via the Akt pathway such as in gastric cancers or promotes autophagy and non-apoptotic cell death thereby sensitising cells to chemotherapy [82]. In oesophageal cancer, miR-193b directly targets the KRAS pathway and thus, as discussed previously for Hummel et al. (2014), its upregulation in the state of cancer would be expected as it exerts negative transcriptional control to halt cellular proliferation [64]. A 2013 study suggested that dosages of ionizing radiation can manipulate the expression profile for miR-193b in some cancers [83], as supported by the results of Chan et al. (2017) [54] and Wen et al. (2016) [53] whereby miR-193b’s expression profile differed with the addition of radiotherapy. Therefore, this could have affected the patient’s expression profiles should their radiation dosages have differed between patients.Despite the conclusions made by Tanaka et al. (2013) [59] and Mahawongkajit et al. (2020) [66] there is little conclusive evidence as to the exact function of miR-200c. The miR-200 family have been shown to be tumour suppressor genes in ovarian cancers [84] in addition to their downregulation upon neoplastic progression of Barrett’s oesophagus [85]. There is additional evidence that miR-200c overexpression may play a role in chemoresistance of oesophageal cancers via also interacting with the Akt pathway [86].Of the 15 articles identified 14 suggested that at least one or more miRNAs could be used to predict response to NAT. In total, 10 of the 15 included studies utilised a pre-determined validated measures of analysing response to treatment therapy, these being the defined tumour regression scales of Mandards, Japan Esophageal Society, Cologne Regression Grade and RECIST 1.0 criteria. The other seven studies used generic measures of response (i.e., Tanaka et al. (2013) [59] utilised “total regression of tumour”). The use of generic means of measuring response reduces the ability to draw comparisons between studies, not only making appraisal more difficult, but reducing validity via decreasing repeatability.Ilhan-Mutlu et al. (2015) [52] is the only study to find no predictive value in any of the miRNAs it analysed, namely miRNA-21 and miR-148a. This is despite the findings previously mentioned by Kurashige et al. (2012) [57] and Komatsu et al. (2016) [60] who found miR-21 was predictive of low response to nChemo in OAC. However, notably with Ilhan-Mutlu et al. (2015) [52] samples used were of SCC histology, which may explain the differing findings. Unlike Ilhan-Mutlu et al. (2015) [52], these studies utilised AUROC to validate the accuracy of predicting and differentiating between responders and non-responders to NAT, further validating their findings and improving external validity. Selection bias was an issue in many of the studies with the most common problems being that of gender bias, small sample sizes or not providing pre-treatment cancer stages for patients. For example, Ilhan-Mutlu et al. (2015) [52] did not provide pre-treatment TNM stages of patients. Thus, concluding whether a lack of response is caused by a cohort of patients entering at a very late stage of presentation (although clinical staging correlates poorly with response) or simply an innate non-response due to other genetic factors, is not possible. Further to this, the breakdown for the number of patients assigned to each chemotherapy regimen does not add up to the total number of patients stated to have taken part in the study (34 in breakdown vs. 36 total participants). Inconsistencies such as these coupled with a relatively narrow age range and small cohort of only one patient responder, reduce validity and likely account for its lack of positive findings. Samples from Kurashige et al. (2012) [57] may have also developed representivity issues because of cohort attrition caused by a fall in participant numbers from 71 to 24. The data set provided on cohort characteristics accounts for the original 71 participants only, therefore it is unknown whether the cohort of patients left over is representative of the general population. However, this was a reasonable change as only 24 serums samples could be obtained both pre- and post-chemotherapy and thus only these samples could be analysed for intra-treatment reductions in miR-21.Although participant characteristics were well classified in Komatsu et al. (2016) [60] they displayed striking similarities to Komatsu et al. (2016) [56], despite not referencing repeated use of cohorts within the methodologies. Not only were the authors almost identical but total number of participants, age ranges and treatment regimens were all identical. The only differences being miR-21 and miR-23a were their studied miRNAs. Repeated use of the same cohorts without expression may be considered poor scientific practice and only reduces how representative data is to the general population. The same issue occurs for Lynam- Lennon et al. (2012) [51] and Lynam–Lennon et al. (2016) [50] as well as Bibby et al. (2015) [49] where participant characteristics match so closely, and the authors carrying out each study, that it is unlikely a different cohort was used. The main issue with these four studies is each reports the most abundant miRNA as being the most predictive; however, utilising the same cohorts should render the same results each time, considering a similar NAT regime was utilised.Chan et al. (2017) [54] and Wen et al. (2016) [53] concluded that miR-193b downregulation was predictive of non-response, yet Niwa et al. (2019) [56] suggested that its downregulation was predictive of response to NAT. The predictive accuracy of the findings in both Chan et al. (2017) [54] and Wen et al. (2016) [53] were significantly higher than Niwa et al. [58] with AUROC values of 0.89 and 0.86 respectively vs. 0.61. This difference may be due to differences in treatment regimens used. Niwa et al. [58] utilised neoadjuvant chemotherapy only (Cisplatin and 5-Fluorouracil) whereas Chan et al. (2017) [54] and Wen et al. (2016) [53] applied neoadjuvant chemoradiotherapy. This could suggest miR-193b has a role in the regulation of chemosensitivity and radiosensitivity due to its differing expression. Chan et al. (2017) [54] also experienced problems separating healthy biopsy tissue and cancerous tissues. As no standardised measure of response (e.g., Mandards) was used and percentage of viable tumour cells present was the only measure used, the presence of large amounts of healthy tissue within the specimen could have skewed results suggesting patients are responding better than they actually are. Further to this, radiation dosages were not specified. Therefore, it is unknown whether the changes in miRNA expression can be accounted for by the type of therapy or because of their role in predicting response. A 2013 study suggested that dosages of ionizing radiation can manipulate the expression profile for miR-193b in some cancers [83]. Therefore, this could have affected the patient’s expression profiles should their radiation dosages have differed between patients.Future research should be carried out in areas considering predicting the response to specific forms of neoadjuvant chemotherapy or chemoradiotherapy in pre-treatment samples. Analysing the differences between drug formulations and dosages in terms of their effects on the predictive power of miRNAs is essential in tailoring treatment to patients [87]. Further to this, more research would be required in producing a score-based model/panel which ranks patients based on multiple miRNAs rather than utilising only one or two miRNAs to attempt to predict response. By producing such a model, it may increase the overall predictive accuracy by including a larger breadth of miRNA expression profiles and thereby increasing the robustness of such a model enabling a quicker translation into clinical usage [45,49].MiR-21 and miR-200c in OSCC patients could prove to be a useful biomarker for predicting a patient’s response to NAT. MiR-193b shows promising rates of predictability and functional applicability in both OAC and OSCC histologies. These three miRNAs appear consistently significant in terms of response prediction and their role in chemotherapy and chemoradiotherapy resistance. Despite this, research is still needed to elucidate their absolute role in a therapeutic response, and crucially, how the timings of therapeutic administration may affect miRNA expression and thus their value in predicting response. The efforts of future research need to focus on understanding the effects NAT regimes have on the predictive value of each miRNA individually, yet links must be formed to produce a multi-miRNA model for accurate prediction of NAT response with clinical utility to allow optimal patient benefit. Given the low response rate to SOC chemotherapy agents and significantly high mortality and morbidity of OAC and OSCC, robust and predictive biomarkers of NAT response are urgently needed in the clinic and must become a research priority.Conceptualization, C.C.J.L. and S.P.B.; methodology C.C.J.L. and S.P.B.; investigation, C.C.J.L. and S.P.B.; resources, C.C.J.L., M.L., S.A., S.R., O.P., T.U. and S.P.B.; writing—original draft preparation, C.C.J.L. and S.P.B.; writing—review and editing, C.C.J.L. and S.P.B., visualization, C.C.J.L. and S.P.B.; supervision, S.P.B.; project administration, C.C.J.L. and S.P.B. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The authors declare no conflict of interest.Current treatment strategies for oesophageal cancer as outlined by the European Society of Medical Oncology. TNM staging: T describes tumour size and any cancer spread into adjacent tissue; N describes cancer spread to adjacent lymph nodes; M describes metastasis (Adapted from “Oesophageal Cancer: ESMO clinical Practice Guidelines” (2016) [27]. Created with BioRender.com, accessed on 17 February 2022).Synthesis and action of miRNA in post-transcriptional gene regulation. Transcription of miRNA gene via RNA Polymerase II forms pri-miRNA, DROSHA (class 2 ribonuclease III enzyme) and DGCR8 cleave the terminal end of the miRNA hairpin to form pre-miRNA. This is exported via RAN and XPO1/5. The miRNA hairpin is then cleaved by Dicer. AGO-2 binds to the double stranded miRNA, unwinds and dissociates the strands then forms a complex with RISC. This leads to either miRNA degradation or inhibition of ribosome binding. Abbreviations: pri-miRNA: primary-miRNA, pre-miRNA: precursor-miRNA, DGCR8: DiGeorge syndrome critical region gene 8, XPO 1/5: Exportin 1/5, RISC: RNA-induced silencing complex. (Created with BioRender.com, accessed on 16 January2022).Schematic illustration of review articles included in this manuscript.Overview of results produced by all 15 discussed articles. The NAT regimen utilised and how this relates to pre-treatment expression profiles in responders to nChemo or nCRT is shown. Abbreviations: OSCC; Oesophageal Squamous Cell Carcinoma, OAC; Oesophageal Adenocarcinoma, nChemo; Neopadjuvant chemotherapy, nCRT; Neoadjuvant Chemoradiotherapy. (Created with BioRender.com, accessed on 17 February 2022).Articles studying the utility of miRNAs in pre-treatment OAC samples to predict response to NAT.Abbreviations: * = Mean age, ↑ = Increased expression, ↓ = Decreased expression, Formalin Fixed -Paraffin Embedded (FFPE), Neoadjuvant Chemoradiotherapy (nCRT), Neoadjuvant Chemotherapy (nChemo), Oesophageal Adenocarcinoma (OAC), Oesophageal Squamous Cell Carcinoma (OSCC), Pathological Complete Response (pCR), Reverse Transcription–Quantitative Polymerase Chain Reaction (RT–qPCR) and Tumour in situ (Tis), Tumour Regression Grade (TRG).Articles studying the utility of miRNAs in pre-treatment OSCC samples to predict response to NAT.Abbreviations: ↑ = Increased expression, ↓ = Decreased expression, Formalin Fixed-Paraffin Embedded (FFPE), Japanese Esophageal Society (JES), Neoadjuvant Chemoradiotherapy (nCRT), Neoadjuvant Chemotherapy (nChemo), Oesophageal Adenocarcinoma (OAC), Oesophageal Squamous Cell Carcinoma (OSCC), Pathological Complete Response (pCR), Response Evaluation Criteria in Solid Tumours (RECIST), Reverse Transcription–Quantitative Polymerase Chain Reaction (RT–qPCR) and Tumour in situ (Tis), Tumour Regression Grade (TRG).Articles studying the utility of miRNAs in pre-treatment OAC and OSCC samples combined to predict response to NAT.Abbreviations: ↑ = Increased expression, ↓ = Decreased expression, Cologne Regression Grade (CRG), Formalin Fixed -Paraffin Embedded (FFPE), Neoadjuvant Chemoradiotherapy (nCRT), Oesophageal Adenocarcinoma (OAC), Oesophageal Squamous Cell Carcinoma (OSCC), Pathological Complete Response (pCR) and Reverse Transcription–Quantitative Polymerase Chain Reaction (RT–qPCR).Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01172.txt

@@ -0,0 +1 @@
+Somatostatin receptors (SSTs) are of particular interest in oncology because these proteins are overexpressed on the cell membranes of different human malignancies, especially neuroendocrine tumors (NETs) and neuroendocrine neoplasms (NENs). Radiolabeled short peptide analogs of the natural hormone somatostatin have been developed over the years to target SST-expressing tumors and are used for both imaging (diagnosis) and therapy. Today, this type of radiopharmaceutical plays a pivotal role in the management of NET and NEN patients. Despite their clinical success, new developments in recent years, in terms of peptide analogs and radionuclides, have shown certain advantages and hold promise for further improvement in both the diagnosis and therapy of SST-expressing tumors, even beyond NETs and NENs.Somatostatin receptors (SSTs) are recognized as favorable molecular targets in neuroendocrine tumors (NETs) and neuroendocrine neoplasms (NENs), with subtype 2 (SST2) being the predominantly and most frequently expressed. PET/CT imaging with 68Ga-labeled SST agonists, e.g., 68Ga-DOTA-TOC (SomaKit TOC®) or 68Ga-DOTA-TATE (NETSPOT®), plays an important role in staging and restaging these tumors and can identify patients who qualify and would potentially benefit from peptide receptor radionuclide therapy (PRRT) with the therapeutic counterparts 177Lu-DOTA-TOC or 177Lu-DOTA-TATE (Lutathera®). This is an important feature of SST targeting, as it allows a personalized treatment approach (theranostic approach). Today, new developments hold promise for enhancing diagnostic accuracy and therapeutic efficacy. Among them, the use of SST2 antagonists, such as JR11 and LM3, has shown certain advantages in improving image sensitivity and tumor radiation dose, and there is evidence that they may find application in other oncological indications beyond NETs and NENs. In addition, PRRT performed with more cytotoxic α-emitters, such as 225Ac, or β- and Auger electrons, such as 161Tb, presents higher efficacy. It remains to be seen if any of these new developments will overpower the established radiolabeled SST analogs and PRRT with β--emitters.The somatostatin family consists of two cyclic disulfide-bond-containing peptide hormones, one with 14 amino acids (SS-14, primary form in the brain) and one with 28 amino acids (SS-28, primary form in the gut). The biologic actions of somatostatin are mediated by five somatostatin receptor subtypes (SST1-5), which belong to a distinct group within the G-protein-coupled receptor superfamily, also known as 7-transmembrane receptors. The activation of these receptors stimulates multiple intracellular cascades to modulate growth hormone release, insulin and glucagon secretion, gastric acid secretion, and neuronal activity. The five subtypes (SST1-5) have approx. 50% identical amino acids, with homology being the most pronounced in the transmembrane regions, and they are subdivided into two subgroups: one consisting of SST2, SST3, and SST5, differing from the other subgroup, which consists of SST1 and SST4 in terms of amino acid homology and pharmacological profile [1]. SSTs are of particular interest in oncology, because their expression is linked to different human malignancies [2,3,4,5]. SSTs are recognized as favorable molecular targets in neuroendocrine tumors (NETs) and neuroendocrine neoplasms (NENs) for targeting and drug delivery, with subtype 2 (SST2) being the predominantly and most frequently expressed [6,7,8].Today, radiopharmaceuticals targeting the SST play a pivotal role in the management of NEN and NET patients [6,7]. These radiopharmaceuticals are mainly based on short peptide analogs of the natural hormone somatostatin, and their clinical success lies in the following factors: (a) the expression of SST in a high incidence and density on the surface of NET cells (easily accessible) compared to their low expression in other tissues; (b) the development, over the years, of synthetic peptide analogs of somatostatin, which have been optimized in terms of in vivo stability, affinity, specificity, and pharmacokinetics; and (c) the advances in radiochemistry and chelation chemistry, which have allowed for the chemical tuning of these peptides for radiolabeling with various radionuclides for different medical applications in nuclear oncology. Undoubtedly, radiolabeled somatostatin analogs have paved the way for a number of modern developments, especially for nuclear oncology and endocrinology. This review features the development and application of SST-targeting radiopharmaceuticals, and it represents both the radiochemist’s and the clinician’s view. This article provides a concise overview of the current status, the latest developments, and the future prospects in the field. More precisely, it presents (I) the radiolabeled SST agonists, including the key structural features of somatostatin that led to the currently established radiopharmaceuticals, their clinical applications, and the most recent advancements; (II) the radiolabeled SST antagonists, from their conceptualization and their structural design in comparison with the agonists to the clinical data and status of their development to date; (III) the current evidence for novel clinical indications of radiolabeled SST analogs, especially antagonists; and (IV) the perspectives of labeling with new radionuclides and of targeting somatostatin receptor subtypes other than SST2.In the amino acid sequence of the endogenous hormone somatostatin, the small tetrapeptide Phe-Trp-Lys-Thr (corresponding to the amino acid residues 7–10 in the natural hormone somatostatin-14) was identified as essential for receptor recognition and biological activity [9,10]. The introduction of d-amino acids for improved in vivo stability and stepwise optimization, based on the minimal amino acid chain length in somatostatin, resulted in an octapeptide with a type II β-turn, formed by the active core Phe-d-Trp-Lys-Thr in a six-member ring via a disulfide bridge, known as octreotide (OC, Table 1) [11]. Octreotide (Sandostatin®) is used for the management of growth-hormone-producing tumors (e.g., acromegaly), and tumor and symptom control of neuroendocrine tumors [12], and it has been the starting point for the development of radiolabeled somatostatin analogs (Figure 1). It is worth mentioning that while the natural hormones somatostatin-14 and somatostatin-28 bind to all subtypes with high (though not the same) affinity, short somatostatin analogs, such as octreotide, only bind to the first subgroup of receptor subtypes (Table 1). More precisely, octreotide has high affinity to SST2 and SST5 and moderate affinity to SST3. The most interesting structural features on octreotide-based analogs are position 3 (Phe), which is involved in the critical β-turn, and position 8 (Thr(ol)), modifications of which have led to analogs with different receptor subtype selectivities and affinities. Briefly, the well-known Tyr3-octreotide (TOC), where Phe is substituted by Tyr, shows high affinity to SST2 and moderate affinity to SST5, while 1-Nal3-octreotide (NOC) and BzThi3-octreotide (BOC) show additional affinity to SST3. The analog with substitution in both positions, [Tyr3, Thr8]-octreotide ([Tyr3]-octreotate or TATE), binds almost selectively to SST2, while the corresponding [1-Nal3, Thr8]-octreotide (NOC-ATE) and [BzThi3, Thr8]-octreotide (BOC-ATE) show additional affinity to SST5 and SST3 [13,14]. See Table 1 for affinity data.After the pioneering work of Lamberts et al. in 1989, where endocrine-related tumors could be visualized using 123I-labeled Tyr3-octreotide (TOC), the conjugation of chelators for labeling with radiometals revolutionized the field (Figure 1) [15]. More specifically, the following advances can be noted: (a) the clinical success of 111In-DTPA-octreotide (OctreoScan®, where DTPA: diethylenetriaminepentaacetic acid); (b) the introduction of the chelator 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid (DOTA), which is able to form thermodynamically and kinetically stable complexes with a series of 3+ radiometals, like the β--emitter 90Y; (c) the introduction of peptide receptor radionuclide therapy (PRRT) with 90Y- or 177Lu-labeled SST agonists, such as 90Y- or 177Lu-DOTA-TOC and 177Lu-DOTA-TATE [16]; and (d) the accelerated development of 68Ga radiochemistry/radiopharmacy, establishing SST PET/CT with somatostatin analogs, such as 68Ga-DOTA-TOC, 68Ga-DOTA-TATE, and 68Ga-DOTA-NOC, allowing the most sensitive staging and restaging of NETs, as well as the identification of patients who would benefit from PRRT (theranostic approach), which made radiolabeled somatostatin analogs the archetype of peptide-based radiopharmaceuticals. Nowadays, a plethora of radiolabeled somatostatin analogs have been developed in order to optimize affinity, specificity, and/or pharmacokinetics (many reviews are available, see, for example, Eychenne R et al. [17]). Among them, DOTA-TOC and DOTA-TATE remain the most widely used analogs, with DOTA-TATE (NETSPOT®) and DOTA-TOC (SomaKit TOC®) kits having approval by the US Food and Drug Administration (FDA) and European Medicines Agency (EMA) for 68Ga-labeling, and 177Lu-DOTA-TATE (177Lu-oxodotreotide or Lutathera®) being the only agent approved for therapy to date. It is expected that the approval of 177Lu-DOTA-TOC (177Lu-edotreotide) will follow the completion of the COMPETE (NCT03049189) phase III trial.Today, PRRT with radiolabeled SST agonists (e.g., DOTA-TOC or DOTA-TATE, Table 1) is part of the standard of care of NENs. NETTER-1 (NCT01578239; EudraCT number 2011-005049-11) was the first prospective, open-label, randomized, phase III trial to compare four cycles of 177Lu-DOTA-TATE (4 × 7.4 GBq) plus 30 mg long-acting release octreotide (PRRT group, n = 117) with high-dose (60 mg double dose) long-acting release octreotide (control group, n = 114) in advanced, progressive midgut NET patients. There was a significantly longer progression-free survival for the PRRT arm (p < 0.001) [22] and a significant improvement in quality of life [23]. Consequently, 177Lu-DOTA-TATE (177Lu-oxodotreotide) received marketing authorization for the treatment of adult patients with SST-positive gastroenteropancreatic neuroendocrine tumors (GEP-NETs). At the final analysis of overall survival (OS), the median OS was improved by 11.7 months for the 177Lu-DOTA-TATE arm versus the control arm (48.0 (95% CI, 37.4–55.2) vs. 36.3 (95% CI, 25.9–51.7) months, respectively), which, however, did not reach statistical significance in the long-term follow-up with a median of 6.3 years [24]. Regarding safety, the NETTER-1 data show a low incidence of long-term side-effects regarding hematotoxicity and nephrotoxicity.Currently, a second prospective, randomized, controlled, open-label, multi-center, phase III trial, COMPETE (NCT03049189), is ongoing, in which PRRT using 177Lu-DOTA-TOC (177Lu-edotreotide, four cycles with 7.5 GBq/cycle) is being compared with the mTOR inhibitor everolimus (10 mg daily) in patients with progressive, SST-positive GEP-NETs. Upon completion of the study, the approval of 177Lu-DOTA-TOC is expected. These trials and other trials (e.g., OCCLURANDOM, NCT02230176) should further precisely determine the position of PRRT in the current clinical algorithm with regard to other systemic therapies, such as everolimus and sunitinib.Using routes other than intravenous administration may be an interesting approach to enhance the therapeutic and safety window of PRRT. NENs and their liver metastases are often highly perfused, and the intra-arterial route can exploit the first-pass effect to treat liver-dominant disease more efficiently. Such an approach can also be used for inoperable primary tumors to downstage the disease in the neoadjuvant setting [25,26]. However, large comparative prospective trials supporting its wider use are missing.Alpha particles have a very short range in tissues (20–100 μm), irradiating volumes with cellular dimensions and therefore sparing normal surrounding tissues from cytotoxic radiation. At the same time, their linear energy transfer (LET) is much higher compared to that of β- particles (50–230 vs. 0.2 keV/μm), which makes alpha radiation far more cytotoxic. Among the α-emitters, 213Bi was initially used in combination with DOTA-TOC. Kratochwil et al. performed the first clinical study (retrospective) with an α-emitter in combination with DOTA-TOC (213Bi-DOTA-TOC) in seven patients with metastatic NETs (activities ranging from 3.3 to 21 GBq in one–five cycles) after progressing under 90Y-/177Lu-DOTA-TOC therapy, and it is already available [27]. The report showed moderate renal and hematological toxicity but possible long-term bone marrow toxicity, with the diagnosis of multiple dysplastic syndrome/acute myeloid lymphoma MDS/AML in one heavily pretreated patient. However, the current supply limitations of high-activity 225Ac/213Bi generators have prevented larger confirmatory prospective studies and have instead motivated the use of the α-emitters 225Ac or 212Pb.212Pb-DOTAMTATE (AlphaMedix™) is in a phase I, non-randomized, open-label, dose-escalation, single-center study in 20 PRRT naïve NET patients (NCT03466216), with the highest dose level being four cycles of 2.50 MBq/kg/cycle. Previously, at the highest dose level in a small cohort of 10 NET patients, the objective radiological response (ORR) was 80%, and it had mild adverse effects and a tolerable safety profile [28].225Ac-DOTA-TOC was administered in 40 patients with progressive NENs, where the maximum tolerated dose was established at 40 MBq as a single fraction and at 25 MBq in two fractions at a 4-month interval [29]. In another study, 225Ac-DOTA-TATE was reported in 32 patients with metastatic GEP-NETs, who were stable or had progressive disease and were on 177Lu-DOTA-TATE therapy. After the administration of 7.8–44.4 MBq 225Ac-DOTA-TATE in one–five portions, partial remission was achieved in 15 patients and stable disease in 9 of them. At a median of 8-month follow-up, no disease progression or deaths were documented [30]. Recently, a retrospective analysis was performed in 39 patients who received 225Ac-DOTA-TOC in an attempt to define the safety levels of 225Ac-DOTA-TOC [31]. The analysis was mainly conclusive regarding acute hematological toxicity but not regarding chronic nephrotoxicity due to pre-existing risk factors. Overall, it was found that a single dose of up to 29 MBq, repeated doses of ~20 MBq in 4-month intervals, and a cumulative dose of 60–80 MBq were hematologically tolerable and avoided high-grade (3/4) hematotoxicity.Although α-emitters offer potential advantages over β--emitters therapeutically, long-term toxicity data are still lacking to properly assess the therapeutic benefit. Importantly, the translocation of radioactive daughter nuclides from the chelator should also be considered as a potential safety hazard for α-emitters with multiple α-emitting daughters, such as 225Ac.Despite the successful outcome of the NETTER-1 study, the objective response rate of patients treated with 177Lu-DOTA-TATE was, at most, 18% [22], probably due to the rapid blood clearance of 177Lu-DOTA-TATE, leading to a suboptimal tumor residence time. An attempt to overcome this drawback was the incorporation of Evans Blue (EB) motifs, which prolongs the half-life of the conjugate in the blood by having low micromolar affinity to albumin. This concept was applied to DOTA-TATE, for which it was shown that treatment with 177Lu-DOTA-EB-TATE was more effective in SST2-expressing xenografts than 177Lu-DOTA-TATE [32,33]. The first dosimetry data of the long-circulating SST2 agonist 177Lu-DOTA-EB-TATE versus 177Lu-DOTA-TATE showed a 7.9-fold increase in tumor dose, which was counterbalanced with an even greater increase in renal and bone marrow absorbed doses [34]. A better response rate (assessed by 68Ga-DOTA-TATE PET/CT) after one cycle of treatment was reported for the EB conjugate [35], but matching doses in the kidneys and bone marrow were not provided. The superiority of 177Lu-DOTA-EB-TATE was not confirmed in an intraindividual comparison versus 177Lu-DOTA-TOC in a limited number (n = 5) of patients [36], where the tumor-to-critical organs’ absorbed dose ratios (defined as therapeutic index) were mainly higher for 177Lu-DOTA-TOC and not for 177Lu-DOTA-EB-TATE. Whether 177Lu-DOTA-EB-TATE has any benefit over the established radiopharmaceuticals is still debatable.The observation that GPCR antagonists may bind to more binding sites than agonists, since their binding is independent of the fraction of receptors coupled to the GTP-binding proteins [37], was the primary reason for the development of radiolabeled SST antagonists. Structurally, the main feature to convert an agonist to an antagonist was shown to be the inversion of chirality at positions 1 and 2 of the octreotide family [38]. From the very first preclinical evaluation, the superiority of radiolabeled SST antagonists over agonists was illustrated in terms of targeting SST-expressing tumors [20]. For example, the first SST2 antagonist 111In-DOTA-BASS (Table 1) showed almost twice higher tumor uptake compared to the agonist 111In-DTPA-TATE, despite its lower affinity (IC50 = 9.4 ± 0.4 nM vs. 1.3 ± 0.2 nM [18,20]), and it also showed binding to a higher number of sites on the cell membrane (Bmax) [20]. This was, further on, confirmed on human tumor tissues by autoradiography when comparing 177Lu-DOTA-BASS with 177Lu-DOTA-TATE [39], and the long-lasting tumor uptake of 177Lu-DOTA-BASS in xenografts in vivo holds promise for therapeutic applications of the antagonists [40]. A series of analogs were developed by systematic substitutions of different amino acids with the aim of identifying the structural features that lead to SST2-selective antagonists with high affinity [41]. The analogs JR11 and LM3 (Table 1) were selected among the ones with the best affinity and highest hydrophilicity, and they were studied in combination with different chelators and various radiometals [21,42].Several reports in the past had shown that adding a radiometal to a chelator–SST agonist conjugate could alter its affinity, with 68Ga systematically improving the SST2 affinity of DOTA-conjugated agonists, as well as their pharmacokinetics, compared to 111In, 90Y, and 177Lu [13,18]. The effect of the radiometal, but also of the chelator, was far more impressive, and even unexpected, for the SST2 antagonists [21,42]. Comprehensive studies with JR11 and LM3 in combination with different chelators, such as DOTA and NODAGA, and various (radio)metals, including Ga, Cu, In, Y, and Lu, have illustrated a very high sensitivity of the SST2 antagonists to the N-terminal modification needed for radiolabeling, and they have shed light on the most promising metal–chelator–antagonist combinations for further development, having the following major impacts: (1) All Ga-DOTA conjugates lost affinity for SST2, contrary to the (radio)metalated In-, Y-, and Lu-DOTA conjugates. The affinity of the Ga-complexes was recovered by replacing DOTA with NODAGA. For instance, 68Ga-NODAGA-LM3 has a 10-fold higher SST2 affinity than 68Ga-DOTA-LM3, and 68Ga-NODAGA-JR11 has an almost 25-fold higher affinity than 68Ga-DOTA-JR11 (Table 1). Therefore, 68Ga-NODAGA conjugates of SST2 antagonists were selected for clinical development. (2) The great potential of using SST2 antagonists became obvious when the low-affinity 68Ga-DOTA-JR11 was compared to 68Ga-DOTA-TATE, which had approx. a 150-fold higher affinity (Table 1). It was found in vivo that 68Ga-DOTA-JR11 outweighed the affinity differences, being even slightly better than the high-affinity 68Ga-DOTA-TATE. Not to mention that the high affinity 68Ga-NODAGA-JR11 was better distinguished than 68Ga-DOTA-TATE in terms of tumor uptake [21].Similarly, the therapeutic counterpart 177Lu-DOTA-JR11 compared to 177Lu-DOTA-TATE showed a higher tumor uptake and, more importantly, a longer tumor residence time, leading to a higher radiation tumor dose [43] and, consequently, delayed tumor growth and longer median survival [44]. The reasons for these observed in vivo differences can be found, at least partially, in the differences between the two radiopharmaceuticals on the cellular level, which were recently investigated [45]. 177Lu-DOTA-JR11 showed faster association, slower dissociation, and longer cellular retention than 177Lu-DOTA-TATE. Despite a comparable high affinity, 177Lu-DOTA-JR11 recognized four times more receptor binding sites than 177Lu-DOTA-TATE. However, more interestingly, while a high excess of antagonist was able to entirely displace the agonist bound on the cell membrane, the agonist could not completely displace the antagonist. Taken together, the antagonist binds not only to additional binding sites but also to different binding sites that are not recognized by the agonist (e.g., uncoupled G proteins) [45]. This observation is clinically relevant, as it indicates that the interruption of somatostatin agonists before treatment with radiolabeled analogs may not be necessary if SST2 antagonists are used.Last but not least, SPECT tracers based on antagonists are missing, but they are also important considering that more than 70% of nuclear medicine procedures still use 99mTc. The first attempts to label SST2 antagonists with 99mTc via the monodentate ligand hydrazinonicotinamide (HYNIC) using ethylenediamine N,N′ diacetic acid (EDDA) as a co-ligand (similarly to the clinically used agonist [99mTc]Tc-HYNIC/EDDA-TOC) failed because the antagonist entirely lost its affinity for SST2 [46], once more depicting the extreme sensitivity of the antagonists to N-terminal modifications. Further studies illustrated that the loss of affinity can be circumvented, to a certain extent, when a spacer of appropriate length and nature (e.g., aminohexanoic acid) is introduced between the antagonist and HYNIC [47]. Nevertheless, the alternative chelating system 6-carboxy-1,4,8,11-tetraazaundecane (N4) seems to be better suited to 99mTc-based SST2 antagonists. In fact, 99mTc-labeled LM3 via N4 ([99mTc]Tc-TECANT-1) has been selected as the first 99mTc-based antagonist for clinical translation [48] under the ERAPerMED project “TECANT” (Ref No. ERAPERMED2018-125). The clinical trial is expected to start soon.The first clinical evidence indicating that imaging with SST2 antagonists may be superior to that with agonists was provided by a prospective study, which included five patients with NETs or thyroid cancer after total-body scintigraphy and a SPECT/CT scan with 111In-DOTA-BASS versus OctreoScan [49]. 111In-DOTA-BASS had a higher tumor detection rate (25/28 lesions) than 111In-DTPA-octreotide (17/28 lesions) in a lesion-based analysis. Meanwhile, based on affinity studies and preclinical results, 68Ga-NODAGA-JR11 (=68Ga-OPS202) was selected for PET/CT imaging studies. Nicolas et al. performed a single-center, prospective, phase I/II study with 12 GEP-NET patients, comparing PET/CT with two micro doses of 68Ga-NODAGA-JR11 (15 and 50 μg/150 MBq) and one micro dose of the potent SST2 agonist 68Ga-DOTA-TOC (NCT02162446). 68Ga-NODAGA-JR11 showed favorable dosimetry results and imaging properties, with the best tumor contrast between 1 and 2 h after injection [50]. 68Ga-NODAGA-JR11 PET/CT showed a significantly higher sensitivity in a lesion-based comparison with 68Ga-DOTA-TOC PET/CT: 93.7% (95% CI: 85.3–97.6%) vs. 59.2% (95% CI: 36.3–79.1%) [51]. In this study, diagnostic efficacy measures were compared against contrast-enhanced CT or MRI. 68Ga-DOTA-JR11 was also assessed clinically, despite its >20 times lower affinity compared to 68Ga-NODAGA-JR11 (Table 1) [21,52,53,54]. Zhu et al. prospectively compared 68Ga-DOTA-JR11 and 68Ga-DOTA-TATE PET/CT in the same patients with NETs [54]. As in the study of Nicolas et al., they detected significantly more liver lesions with the SST2 antagonist (552 vs. 365) but, at the same time, significantly less bone lesions (158 vs. 388) compared to 68Ga-DOTA-TATE. Importantly, 68Ga-DOTA-JR11 showed a lower tumor uptake than 68Ga-DOTA-TATE, which is in contrast to the study of Nicolas et al., who prospectively compared 68Ga-NODAGA-JR11 and 68Ga-DOTA-TOC PET/CT in the same patients [51]. This finding can be explained by the much lower SST2 affinity of 68Ga-DOTA-JR11 in comparison to 68Ga-NODAGA-JR11 (Table 1) and/or by the study design, which may have caused a bias, as 68Ga-DOTA-TATE PET/CT was always performed 24 h ahead of 68Ga-DOTA-JR11 PET/CT, creating the risk of receptor occupation and/or internalization [55].The therapeutic companion 177Lu-DOTA-JR11 (=177Lu-OPS201), which was initially assessed in a single-center, prospective, proof-of-principle study (phase 0 study), was compared with 177Lu-DOTA-TATE in the same four patients with advanced, metastatic neuroendocrine neoplasia (NEN) (grades 1–3) [56]. The median tumor dose was 3.5-fold higher for the antagonist. At the same time, tumor-to-kidney dose ratios were >2-fold higher with 177Lu-DOTA-JR11 compared to 177Lu-DOTA-TATE. Overall, tumor doses with 177Lu-DOTA-JR11 were up to 487 Gy, with moderate adverse events with grade 3 thrombocytopenia after treatment with three cycles (total 15.2 GBq) in one patient. Figure 2 illustrates a direct comparison of the antagonist 177Lu-DOTA-JR11 versus the agonist 177Lu-DOTA-TOC in the same patient with lung NETs (G2).Later on, a single-center phase I study with 20 NET patients (grades 1–3) reported a best overall response (RECIST 1.1 criteria) of 45%, and the median progression-free survival (PFS) was 21 months (95% CI, 13.6-NR), accompanied, however, with grade 4 hematotoxicity (leukopenia, neutropenia, and thrombocytopenia) in four out of seven patients treated with two cycles of 177Lu-DOTA-JR11 (cumulative activity between 10.5 and 14.7 GBq) [57]. Hence, the study was suspended, and the protocol was modified to limit the cumulative absorbed bone marrow dose. 177Lu-DOTA-JR11 (177Lu-OPS201) is currently being evaluated in a phase I/II, multi-center, open-label study (NCT02592707—active, not recruiting). To date, there is only an abstract available with a brief summary of the results of 20 NET patients with an adequate follow-up [58]. The disease control rate (DCR) at 12 months was 90% (95% CI: 68.3–98.8) for these 20 patients.More recently, in parallel to the development of the theranostic pair based on JR11, the other antagonist, LM3, was also developed. Results in the form of abstracts have reported the feasibility of PET/CT imaging with 68Ga-NODAGA-LM3 in 40 patients with GEP-NET, lung NET, paraganglioma/pheochromocytoma, etc. [59], and a higher detection rate of 68Ga-NODAGA-LM3 versus 68Ga-DOTA-TOC PET/CT in 10 paraganglioma patients, with 68Ga-NODAGA-LM3 PET/CT detecting many more lesions (243 vs. 177), including bone lesions (190 vs. 143) [60]. Meanwhile, 68Ga-NODAGA-LM3 and 68Ga-DOTA-LM3 were compared in a randomized, double-blind study with 16 NET patients [61]. The SUVmax values of tumors and SST2-positive organs were >2 times higher with 68Ga-NODAGA-LM3 than with 68Ga-DOTA-LM3 at 2 h post-injection, which is consistent with the almost 10 times higher SST2 affinity of 68Ga-NODAGA-LM3 compared to 68Ga-DOTA-LM3 (Table 1) [21].The therapeutic companion, 177Lu-DOTA-LM3, was evaluated in a single-center, compassionate-use study, which included 51 patients with metastatic NENs of grades 1–3, who were selected after 68Ga-NODAGA-LM3 PET/CT imaging [62]. There were few adverse events (maximal grade 3 thrombocytopenia in 5.9% of patients) after treatment with one–four cycles of 177Lu-DOTA-LM3, with mean cumulative activity between 6.1 and 26.1 GBq. The partial response and DCR (RECIST 1.1 criteria in 47 patients) were 36% and 85% at 3–6 months, respectively.The binding capacities of radiolabeled SST antagonists and agonists were compared in human tissue samples from nine different tumors using in vitro autoradiography with 177Lu-DOTA-BASS vs. 177Lu-DOTA-TATE [39], as mentioned above, and with 125I-JR11 vs. 125I-TOC [63]. A summary of the outcome is provided in Figure 3.In all tested cases, the radiolabeled SST2 antagonist bound to more SST2 sites in all tumors, with an uptake that was 3.8–21.8 times higher than that with the agonist. Interestingly, in some non-neuroendocrine neoplasias, the level of binding of the antagonists reached the same level as that of the agonists (e.g., 177Lu-DOTA-TATE) in well-differentiated NENs. Of particular interest is the fact that tumors other than GEP-NETs and lung NETs have the potential to become targets for radiolabeled SST2 antagonists, despite the relatively low SST2 expression, for example, non-Hodgkin lymphomas, renal cell carcinoma, breast cancer, pheochromocytoma, paraganglioma, medullary thyroid cancer, small-cell lung cancer, and paraganglioma.SSTs are also expressed in peritumoral vessel endothelial cells; in inflammatory cells; and in immune system cells, such as activated lymphocytes, monocytes, and epithelioid cells. This suggests that clinical indications can be found in benign and chronic inflammatory diseases, besides oncology [64]. PET imaging with 68Ga-labeled SST agonists (DOTA-TOC, DOTA-TATE, and DOTA-NOC) have shown relevance in detecting vulnerable, atherosclerotic plaques and have been correlated to other risk factors in patients (summarized in [65]). The use of antagonists in this context has, to date, only been explored preclinically [66]. In terms of PRRT, a retrospective analysis of a limited number of oncological patients indicated that 177Lu-DOTA-TATE results in a reduction in atherosclerotic plaque activity [67], while 177Lu-DOTA-TOC showed treatment effects in a feasibility study involving two patients with refractory multi-organ involvement of sarcoidosis [68]. Nevertheless, radiolabeled somatostatin analogs have not yet found clinical relevance in these indications, and their impact on clinical outcome needs to be assessed in large-scale clinical trials.The use of alternative theranostic pairs of radionuclides, such as radioisotopes of scandium (43/44/47Sc) and terbium (149/152/155/161Tb), might open novel theranostic applications. Recently, a preclinical study demonstrated clear therapeutic benefit when using 161Tb instead of 177Lu in combination with SST analogs; 161Tb has similar decay properties to 177Lu but, additionally, emits a substantial number of conversion and Auger electrons [69]. The most important finding of the study was the identification of the cellular localization of the 161Tb-labeled SST analog, which leads to the best therapeutic outcome. It was shown that the combination of 161Tb with the SST2 antagonist DOTA-LM3, which is not internalizing but remains on the cell membrane, was a better combination than the internalized cytoplasm agonist DOTA-TOC and the internalized and partially nucleus-localized DOTA-TOC-NLS bearing a nucleus-targeting unit (nuclear localization signal (NLS)) [69]. Overall, the preclinical data suggest a benefit of treating NENs with 161Tb-DOTA-LM3 (or 161Tb-labeled SST antagonists) vs. 177Lu-DOTA-TOC (161Tb-labeled SST agonists).To date, radiolabeled somatostatin analogs used for treatment bind with high affinity to the most predominantly expressed SST2. However, various expression and co-expression patterns have been described for the five somatostatin receptor subtypes (SST1-5), depending on the tumor type and origin [3,70,71]. Interestingly, tumor areas lacking expression of a given subtype may be populated by another one [70,71]. In addition, the downregulation or loss of SST2 in advanced disease stages is associated with an inherently worse disease prognosis, a lower sensitivity in imaging, and ineffective therapy with SST2-specific analogs due to inadequate tumor targeting. Hence, somatostatin analogs with affinity to more than one receptor subtype are of great interest, as they address receptor subtype co-expression and heterogeneous expression patterns [72]. Analogs targeting more subtypes than SST2 potentially target a broader spectrum of tumors and/or increase the uptake of a given tumor and are, therefore, a field to explore.Conceptualization: M.F. and R.M.; writing—original draft preparation, M.F., R.M., G.P.N. and D.W.; writing—review and editing, M.F., R.M., G.P.N. and D.W. All authors have read and agreed to the published version of the manuscript.This research received no external funding.D.W. reports personal fees from Ipsen and grants form Siemens Healthineers.Evolution in the development of radiolabeled somatostatin analogs. Color code: orange for L-amino acids (also showing the essential amino acids (tetrapeptide) in the somatostatin sequence for receptor recognition), green for D-amino acids, blue for chelators, and yellow for radionuclides.Multi-intensity projection SPECT of 177Lu-DOTA-JR11 and 177Lu-DOTA-TOC in the same patient with metastatic atypical lung carcinoid. Arrows show tumor doses of 177Lu-DOTA-JR11 vs. 177Lu-DOTA-TOC: red arrow: 12.6 vs. 3.36 Gy/GBq, blue arrow: 9.89 vs. 1.46 Gy/GBq.Radiolabeled SST2 agonist/antagonist binding in different human tumors. 125I-JR11/125I-Tyr3-octreotide data are from [63]. 177Lu-DOTA-BASS/177Lu-DOTA-TATE data are from [39]. Numbers indicate the samples sizes.Somatostatin-based radiotracers. Affinity data (IC50 = half maximal inhibitory concentration) and clinical status.1-Nal = 1-naphthyl-alanine; Aph(Hor) = 4-amino-L-hydroorotyl-phenylalanine; D-Aph(Cbm) = D-4-amino-carbamoyl-phenylalanine. n.r. = not reported. * Data are from [18]; # Data are from [19] (different lab); & Data are from [13]; ¥ Data are from [20]; ¶ Data are from [21].Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01173.txt

@@ -0,0 +1 @@
+These authors contributed equally to this work.In hematological neoplasms associated with COVID-19, immunological dysfunction, including reduced count of non-classical monocytes, has been suggested as a primary driver of morbidity and mortality. In this work, we investigated the contribution of absolute monocyte count to clinical outcome of COVID-19 in 120 patients affected by hematological neoplasms that tested positive to SARS-CoV-2. We found that there was no statistical difference in 30-day mortality, rate of hospitalization for intensive cure and viral clearance at 14 days between fully vaccinated and unvaccinated patients. Increased 30-day mortality was associated with presence of active/progressing disease and absolute monocyte count lower than 400 cells/uL. Reduced absolute counts of monocytes should be used as an alert of increased risk of severe/critical forms of COVID-19 in patients with hematological malignancies, even when the full vaccination cycle has been completed.Background: Clinical course of COVID-19 depends on several patient-specific risk factors, including immune function, that is largely compromised in cancer patients. Methods: We prospectively evaluated 120 adult consecutive patients (including 34 cases of COVID-19 breakthrough after two full doses of BNT162b2 vaccine) with underlying hematological malignancies and a SARS-CoV-2 infection, in terms of patient’s clinical outcome. Results: Among fully vaccinated patients the achievement of viral clearance by day 14 was more frequent than in unvaccinated patients. Increased 30-day mortality was associated with presence of active/progressing disease and absolute monocyte count lower than 400 cells/uL. Results of multivariable analysis in unvaccinated patients showed that the pre-infection absolute count of monocytes less or equal to 400 cells/mmc, active or progressive disease of the underlying hematological malignancy, the COVID-19 severity identified by hospitalization requirement and lack of viral clearance at 14 days were independent predictors of 1-year overall survival. Conclusions: Taken together, our results indicate that absolute monocyte count determined one month before any documented SARS-CoV-2 infection could identify patients affected by hematological neoplasms with increased risk of inferior overall survival.COVID-19 (coronavirus disease-2019) is a complex disease with variable clinical presentations and outcomes, due to the infection of a novel β-Coronavirus SARS-CoV-2 [1]. In most cases, COVID-19 symptoms are moderate or totally absent, with about one week of incubation period. Around 15% of patients can progress to severe pneumonia and about 5% eventually progress to acute respiratory distress pneumonia, renal failure, septic shock, multiple organ failure and death [2,3,4]. Due to the heterogeneous clinical course of COVID-19, several biomarkers have been evaluated that could allow us to predict an initial severe presentation or critical evolution of the disease. Several observational studies have suggested that some comorbidities, assessed by high Charlson comorbidity index (CCI ≥ 3) scores are disproportionately associated with inferior clinical outcomes [5].Clinical course and disease severity in COVID-19 is strongly associated with weaker immune response, bulk release of proinflammatory cytokines and the recruitment of neutrophils, monocytes and macrophages, which can generate an aggressive response, in some cases inappropriate, detrimental and harmful for the host [6], especially in cancer patients [7,8]. Recent reports disclosed the contribution of monocytes to the hyper-inflammatory phenotype, thus worsening disease severity [9,10,11]. In vitro, SARS-CoV-2 can induce a functional specialization of dendritic cells subsets, with high levels of interferon-α, IL-6, IL-10 and IL-8 that orchestrate and propagate first the innate and then the adaptive immune response [8,12,13]. High levels of IL-6 can further trigger cytokine storm in the absence of appropriate type I and III interferon response [14,15]. Patients carrying hematological neoplasms have increased infection susceptibility, due to immunodeficiency, T-cell anergy, increased myeloid-derived suppressor cells and impairment of antigen presentation machinery [6,13,16], as a consequence of the malignancy itself. In multiple myeloma (MM), either anergic, dysfunctional effector lymphocytes or both, tumor-educated myeloid-derived suppressor cells and soluble mediators promote coordinately cancer immune-evasion [17,18,19,20,21,22]. In chronic lymphatic leukemia (CLL), circulating monocytes have a skewed aberrant phenotype leading to altered composition and phagocytosis, contributing T-cell exhaustion and anergy [23,24,25,26]. In Hodgkin’s lymphoma (HL) monocyte-like myeloid derived suppressor cells are increased and correlate to chemosensitivity [19,27,28,29,30,31]. In acute myeloid leukemia (AML), macrophages and myeloid-derived cells are educated by leukemia itself to develop a supportive phenotype, and thus, contribute to drug resistance [32,33,34,35].However, little is known about the contribution of perturbed counts and functions of monocytes to COVID-19 clinical course in patients carrying hematological malignancies, and if the perturbed immune parameters observed in critically ill COVID-19 patients retain the same significance in the setting of patients affected by hematological neoplasms. To this end, we designed a single-center prospective study, to determine the contribution of monocyte accounting in the clinical outcome of 120 consecutive patients affected by hematological neoplasms and tested positive to SARS-CoV-2 in our center from 15 April 2020 through 30 November 2021. Our study included 120 adult consecutive patients (aged ≥18 years) with a WHO-defined hematologic malignancy that tested positive to SARS-CoV-2 in the emergency departments, hospital wards (patients infected while hospitalized) or outpatient clinics of the Division of Hematology, AOU Policlinico in Catania, Sicily, Italy from 15 April 2020 through 30 November 2021. Patients were categorized as fully vaccinated at the time of COVID-19 when two doses of vaccine BNT162b2 had been administered and diagnosis of COVID-19 was recorded >4 weeks from the last dose, thus identifying the breakthrough infections after COVID-19 vaccination. Unvaccinated patients were defined as having no known prior exposure to COVID-19 vaccination before COVID-19 diagnosis.The COVID-19 diagnosis was confirmed by nasopharyngeal swab collection in accordance with local prevention guidelines. The study was approved by the Institutional Review Board (IRB) (Comitato etico Catania 1, https://www.policlinicovittorioemanuele.it/comitato-etico-catania-1 (accessed on 15 April 2020), #CO.TIP. 34/2020/PO 0,016,693 released on 15 April 2020) and performed in accordance with the principles of the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines.Electronic health records of patients followed in our hospital were evaluable to capture the following information: type of hematological malignancies, details and timing of the tumor treatment, laboratory parameters at the time of infection, outcome of the SARS-CoV-2 infection and outcome of the hematological malignancies at the time of last follow-up. Active antineoplastic treatment was defined as having received anticancer therapy within 30 days prior to COVID-19 diagnosis. Infection laboratory values were collected no more than 7 days preceding the first documentation of SARS-CoV-2 infection. COVID-19-related death was categorized in accordance with the WHO definition, while comorbidity was classified according to the modified Charlson comorbidity index (CCI).In unvaccinated patients, seroconversion was performed on serum samples to detect human antibodies of the immunoglobulin classes IgG and IgA against the SARS-CoV-2 by anti-SARS-CoV-2 ELISA IgG and IgA assays (Euroimmun), until one month from documented viral clearance, according to the manufacturer’s instructions. In vaccinated patients, the titer of antibodies developed against the receptor-binding domain of the SARS-CoV-2 spike protein was measured 30 days after the second dose of BNT162b2 vaccine using the SARS-CoV-2 IgG II Quant assay (Abbott, CE marked), by chemiluminescence (CMIA) method, performed on the Abbott Alinity i platform according to the manufacturer’s instructions. Continuous variables were expressed as median and range (minimum–maximum), since a preliminary analysis showed that data distribution was not normal. Normality was verified using the Shapiro–Wilk test and graphically using Q–Q plot. Counts and percentages of qualitative variables were generated for descriptive statistical analysis. For further comparisons, we used the Mann–Whitney U test for continuous data and the Fisher’s exact test for categorical data. We applied propensity score matching (PSM) as a consequence of the limited sample size and its heterogeneity to adjust for differences in baseline clinical variables between fully vaccinated and unvaccinated patients [8]. The covariates balanced between groups were: age (used as a dichotomic variable, less than 70 years, equal to or more than 70 years old), biologic sex (female; male), Eastern Cooperative Oncology Group performance status (ECOG PS 0–1; ≥2), lymphopenia (absolute lymphocyte count (ALC), ≤1000 vs. >1000 per μL), Charlson comorbidity index (0–1; vs. ≥2), cancer status (active and progressing vs. not active and progressing), hematological neoplasm type (lymphoid; myeloid or plasma cell neoplasm). In PSM, we selected the caliper 0.25 of the standard deviation for drawing the control units (unvaccinated) to match the treated units (fully vaccinated) with the nearest-neighbor method and a 3:1 ratio (unvaccinated:fully vaccinated). Due to limited number of events, we considered variable selection in regression analysis by elastic-net regularization with a mixing parameter (LASSO). Following LASSO variable selection, the additional inclusion of AMC < 400 cells/uL was considered given a significant association with the primary endpoint and prior evidence suggesting the association of this variable to immune dysregulation in COVID-19 patients carrying hematological neoplasms [6]. The primary endpoint was 30-day all-cause mortality (infection, progressive disease, other) among fully vaccinated patients affected by hematologic malignancy who tested positive to SARS-CoV-2 compared to the cohort of unvaccinated hematological patients after PSM adjustment for baseline clinical variables. Secondary endpoints included rates of hospitalization in intensive care units and viral clearance at 14 days, in fully vaccinated, compared with unvaccinated patients with hematological neoplasms after PSM adjustment for baseline clinical variables.All calculations were performed using MedCalc Statistical Software version 13.0.6 (MedCalc Software bvba, Ostend, Belgium; http://www.medcalc.org (accessed on 6 January 2022); and XLSTAT version 2021.5-Life Sciences, released in December 2021.We diagnosed COVID-19 infection in 120 patients affected by hematological neoplasms, of whom 86 (72%) were unvaccinated and 34 (28%) were fully vaccinated, including 18/34 (15%) patients who had received a boost dose within 2 weeks from COVID-19 diagnosis. Baseline characteristics of the cohort are summarized in Table 1. The median age was 65 years (range 23–94 years) in the unvaccinated group, without any significant difference to fully vaccinated patients. In the fully vaccinated group, most patients were receiving active treatment for their hematological malignancy, differently from the unvaccinated group (p < 0.0001), mostly consisting of immunotherapy and targeted therapy, like lenalidomide maintenance in patients affected by multiple myeloma. The median value of Charlson comorbidity index was 3 (0–14), in the unvaccinated group, with the highest score among patients affected by lymphoid neoplasms, due to cardiovascular disease, pulmonary disease and diabetes (data not shown). A Charlson comorbidity index equal to or higher than 2 was more frequent in the unvaccinated group than in vaccinated patients (p = 0.003, Table 1).There were no significant differences in pre-infection laboratory parameters evaluated among unvaccinated and fully vaccinated patients, including absolute counts of neutrophils, monocytes and lymphocytes, C-reactive protein (C-RP) and lactate de-hydrogenase, which have been associated with COVID-19 severity in both general population and cancer subjects [36,37].Among fully vaccinated patients the achievement of viral clearance by day 14 was more frequent than in unvaccinated patients (p = 0.0003, Table 2). Only one fully vaccinated vs. 22 (26%) unvaccinated patients was admitted for inpatient intensive care (p = 0.004, Table 2), due to concomitant neutropenic fever and bacterial pneumonia. Unvaccinated patients who achieved viral clearance by 14 days from the documented SARS-CoV-2 infection had a lower Charlson comorbidity score, due to significant lower frequency of heart disease (32%, p = 0.04, data not shown). There was no significant difference among patients affected by myeloid, lymphoid or plasma cell neoplasm, active or not (data not shown). However, the number of patients who died from COVID-19 could not reach any significant difference among fully and unvaccinated patients (Table 2). Unvaccinated patients who required hospitalization for their COVID-19 in an intensive care unit were older than those who did not require hospitalization (p = 0.002), with higher median Charlson comorbidity index (5 vs. 3), due to significantly higher frequency of heart disease (p = 0.01). There was no significant difference among patients affected by myeloid, lymphoid or plasma cell neoplasm, active or not. However, there were more patients in treatment with chemotherapy among hospitalized patients (p = 0.04), while non-hospitalized patients received targeted therapy more frequently (p = 0.004), reflecting the higher probability of hospitalization for those with chemotherapy-related immune deficiency. Patients requiring hospitalization for COVID-19 had lower hemoglobin (p = 0.03) and absolute monocyte counts (p = 0.02, Table 3).The median value of anti-SARS-CoV-2 antibodies (IgG) titer was 4.7 (range 1.1–38.7 BAU), as measured at one month from the documented viral clearance in 66/86 unvaccinated patients, showing that 42/66 (64%) patients were seroconverted.Among the fully vaccinated patients, the anti-SARS-CoV-2 antibodies (IgG) titer was 5.2 (range 0–13.1 BAU) in 24/34 tested patients, showing that none of them achieved a protective titer (>40 BAU) at one month after two doses of BNT162b2 vaccine. Following PSM there was no statistical difference in 30-day mortality, rate of hospitalization for intensive cure and viral clearance at 14 days between fully vaccinated and unvaccinated patients, as shown in Table 4 where adjusted odds ratios (AOR) and 95% confidence intervals (CI) have been reported. Increased 30-day mortality was associated with the presence of active/progressing disease and absolute monocyte count lower than 400 cells/uL (presence of non-active disease vs. active/progressing disease: AOR −0.64, 95% CI: −1.09–0.2; absolute monocyte count (AMC) less than 400 cells/uL vs. AMC ≥ 400 cells/uL, AOR 0.68, 95% CI: 0.14–1.21). We then assessed potential biomarkers for 1-year overall survival (OS) of patients carrying hematological neoplasms that tested positive for SARS-CoV-2 infection. To this end, we considered only the cohort of unvaccinated patients due to their longer median follow-up (13.6 months).In unvaccinated patients of our series, the median OS was 10.4 months (95% CI. 9.6–11.3), which was affected in univariate analysis, summarized in Table 5, by presence of: active disease (p = 0.003), pre-infection absolute count of monocytes less or equal to 400 cells/mmc (p = 0.04), hospitalization due to COVID-19 (p < 0.0001), lack of viral clearance at 14 days (p = 0.005) and lack of seroconversion (p = 0.04). The difference in survival was consistent throughout all subgroups tested, independent of whether the patients were aged below or above 70 years, of female or male sex, or suffered from myeloid, lymphoid or plasma cell neoplasm. Results of multivariable analysis of OS showed that the pre-infection absolute count of monocytes less or equal to 400 cells/mmc (p = 0.008), active or progressive disease of the underlying hematological malignancy (p = 0.009), COVID-19 severity identified by hospitalization requirement (p = 0.004) and lack of viral clearance at 14 days (p = 0.03) were independent predictors of 1-year OS in unvaccinated patients affected by hematological neoplasms tested positive for SARS-CoV-2 infection (Table 6).Several studies showed higher risk of death in cancer patients with COVID-19, along with higher rates of admission in the intensive care unit and development of severe complications from COVID-19 [4,37,38,39,40,41,42,43,44]. Worst outcomes, with death rates of 18.18% and 33.33%, were registered respectively among lung and blood cancer patients [45,46]. In our series of 120 patients affected by hematological neoplasms, the outcome was poor, with higher frequency of hospitalization and death compared to data obtained from general settings, confirming the first report published in August 2020, based on a multi-center series of 536 hematological patients with COVID-19 [46]. However, breakthrough COVID-19, occurring in those patients who developed COVID-19 despite full vaccination cycle in our series was mild, being associated with a lower rate of hospitalization and higher frequency of early viral clearance at 14 days. Preliminary reports show that patients with hematological neoplasms have increased risk of developing breakthrough COVID-19 following full vaccination and remain susceptible to severe outcomes [47].In patients with severe COVID-19, significant and global alterations in both T-, B- and myeloid cell compartments have been described, and this underlines the immune system’s effort to make up for lymphopenia and loss of naïve T-cells by recruiting switched B-memory cells, besides the cytokine storm and the functional and phenotypic alterations of the innate response compartment [6,13,48].The inappropriate response to the virus in patients with moderate and severe disease is the result of a complex interaction between virus, host and environment which affect entry, replication, egress and innate immune control [9,49]. In our series we could not find any correlation between clinical severity or overall survival after COVID-19 associated to changes in absolute counts of neutrophils. Differently from data obtained from the general population or in cancer patients, the pre-infection absolute count of neutrophils and lymphocytes, C-RP and LDH were not associated to OS in unvaccinated patients included in our series [36,46]. However, C-RP and LDH were higher in those patients who did not achieve viral clearance by 14 days, reflecting the impairment of the inflammatory status in COVID-19 [50]. Surprisingly, patients with inferior outcome after COVID-19 had lower absolute monocyte count, differently from previous reports in the general population, where an impairment in monocyte counts and function has been largely described [9,10,51,52]. Indeed, large studies on the general population showed that patients in severe-stages of COVID-19 had increased amounts of circulating CD14+ CD16+ monocytes which exert inflammatory activity through increased release of IL-6 and interaction with adaptive B and T-cells [53,54]. However, decreased frequencies of non-classical monocytes, as consequence of dysregulated emergency myelopoiesis, has been proposed to discriminate patients who develop a severe form of COVID-19 [55,56]. Compared to COVID-19 patients without hematological cancer, patients carrying hematological neoplasms have decreased percentages of classical monocytes, immunoregulatory natural killer cells, double-positive T cells and B cells, that could compromise an initial response to the infection [6].Despite the limited number of patients involved in our work, our data confirm the impairment in the viral clearance in patients affected by hematological neoplasms, and the importance of early viral clearance on clinical outcome and overall survival. Patients who did not develop seroconversion after COVID-19 vaccination had higher probability of achieving viral clearance by 14 days, an independent predictor of overall survival at one year for unvaccinated patients [2,50,57].Thus, if on one hand patients carrying hematological neoplasms had increased risk of not developing a serological response against COVID-19 infection [58,59] or vaccine [60,61,62,63,64], due to treatment-mediated immune dysfunction, on the other hand they can improve viral clearance if fully vaccinated [65,66]. This hypothesis needs to be tested in larger, multi-center cohorts, to explore the clinical course of COVID-19 breakthroughs [47] and identify an effective prevention strategy in contrasting severe/critical forms of COVID-19 in patients with hematological malignancies.Taken together, our results indicate that absolute monocyte count determined one month before any documented SARS-CoV-2 infection could identify among patients affected by hematological neoplasms those with increased risk of inferior overall survival at one year, due to increased risk of hospitalization, lack of seroconversion and viral clearance at 14 days.All authors have made substantial contributions to all of the following: Project administration, A.R. and C.C. (Claudio Cerchione); methodology: A.R., C.C. (Claudio Cerchione) and F.D.R.; A.R. and C.C. (Claudio Cerchione), interpreted the data and drafted the final article; C.C. (Claudio Cerchione) and A.R., performed statistical analysis; A.R., C.C. (Claudio Cerchione), C.C. (Concetta Conticello), S.F., A.B., A.C., V.D.F., S.L., U.M., G.M., M.P. and F.S., selected patients, acquired, analyzed and interpreted the data; F.D.R., G.A.P., C.C. (Concetta Conticello) and C.C. (Claudio Cerchione), revised the article for important intellectual content and approved the final version for submission. All authors contributed to the article and approved the submitted version. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The study was conducted in accordance with the Declaration of Helsinki, and approved by the Ethics Committee of Azienda Policlinico Rodolico San Marco (Comitato etico Catania 1, protocol code #CO.TIP. 34/2020/PO 0016693, date of approval: 15 April 2020).Informed consent was obtained from all subjects involved in the study.The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding authors.F.D.R. and A.R. received honoraria by Amgen, Janssen, Celgene and Takeda. GAP received honoraria from Abbvie, AOP, AstraZeneca, BMS Celgene, Jannsen, Novartis. U.M. received honoraria from Amgen. The other authors declare no competing financial interests. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.Pre-infection clinical and laboratory parameters in hematological patients affected by COVID-19 by baseline vaccination status.a based on Fisher-exact test, b based on Mann–Whitney test. Abbreviations: ALC, absolute lymphocyte count; AMC, absolute monocyte count; ANC, absolute neutrophil count; uL, micro-liter; IU/L, international units per liter; LDH, lactate de-hydrogenase; C-RP, C-reactive protein; ECOG, Eastern Cooperative Oncology Group.Clinical outcomes of COVID-19 in hematological patients by baseline vaccination status.a based on Fisher-exact test.Pre-infection clinical and laboratory parameters in hematological patients hospitalized or not hospitalized for concomitant COVID-19.a based on Fisher-exact test, b based on Mann–Whitney test. Results are reported as median and interquartile range (IQR). Abbreviations: Hb, hemoglobin; ALC, absolute lymphocyte count; AMC, absolute monocyte count; ANC, absolute neutrophil count; uL, micro-liter; IU/L, international units per liter; LDH, lactate de-hydrogenase; C-RP, C-reactive protein; NA, not available.Results of regression analysis of clinical outcomes of COVID-19 in hematological patients following PSM.Abbreviations: LDH, lactate de-hydrogenase; C-RP, C-reactive protein; AMC, absolute monocyte count, uL, microliter, IU/L, international units per liter.Univariate analysis of median overall survival based on main pre-infection clinical and laboratory parameters in unvaccinated hematological patients with COVID-19.a based on Kaplan–Meier plots, p-value was considered significant if <0.05 and indicated in bold italic. Abbreviations: ALC, absolute lymphocyte count; AMC, absolute monocyte count; ANC, absolute neutrophil count; uL, micro-liter; IU/L, international units per liter; LDH, lactate de-hydrogenase; C-RP, C-reactive protein. * is to highlight that only 66 patients were evaluated for seroconversion.Multivariable analysis of median overall survival based on main pre-infection clinical and laboratory parameters found significant in univariate analysis in hematological patients with COVID-19.a hazard ratios and 95% confidence intervals were estimated with Cox regression analysis. Abbreviations: uL, microliter, HR, hazard ratio; CI, confidence interval.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01174.txt

@@ -0,0 +1 @@
+Cancer relapses after chemo-radiotherapies arise from cancer stem cells able to escape cell killing because of their high antioxidants level. The aim of this study was to test the efficacy of ozonized oils to decrease the rate of cancer relapses. In vitro, oils at high ozonide content penetrate inside cancer cells releasing oxygen and reactive oxygen species damaging the thin outer membrane of inactive mitochondria. This event triggers intracellular calcium release and activates apoptosis. In vivo, ozonized oil has been administered by the oral route effectively decreasing blood antioxidants in cancer patients. This approach results in significant increase of survival rate and decrease of relapses in 115 cancer patients (brain, lung, pancreas, colon, skin) undergoing standard radio-chemotherapy regimens during a 4-years follow up. Obtained results indicate that the administration of ozonized oil represents an integrated approach to decrease the risk of radio-chemoresistance and cancer relapses in cancer patients.Background: Cancer tissue is characterized by low oxygen availability triggering neo angiogenesis and metastatisation. Accordingly, oxidation is a possible strategy for counteracting cancer progression and relapses. Previous studies used ozone gas, administered by invasive methods, both in experimental animals and clinical studies, transiently decreasing cancer growth. This study evaluated the effect of ozonized oils (administered either topically or orally) on cancer, exploring triggered molecular mechanisms. Methods: In vitro, in lung and glioblastoma cancer cells, ozonized oils having a high ozonide content suppressed cancer cell viability by triggering mitochondrial damage, intracellular calcium release, and apoptosis. In vivo, a total of 115 cancer patients (age 58 ± 14 years; 44 males, 71 females) were treated with ozonized oil as complementary therapy in addition to standard chemo/radio therapeutic regimens for up to 4 years. Results: Cancer diagnoses were brain glioblastoma, pancreas adenocarcinoma, skin epithelioma, lung cancer (small and non-small cell lung cancer), colon adenocarcinoma, breast cancer, prostate adenocarcinoma. Survival rate was significantly improved in cancer patients receiving HOO as integrative therapy as compared with those receiving standard treatment only. Conclusions: These results indicate that ozonized oils at high ozonide may represent an innovation in complementary cancer therapy worthy of further clinical studies.Cancer cells differ from normal cells in several aspects, among which the blockage of the mitochondrial function stands out. Mitochondrion is the main endogenous source of oxidizing molecules. The mitochondrial blockage occurring in cancer cells is known as the Warburg effect [1]. Recent experimental research provided evidence that cancer cells are favored in their growth by antioxidant molecules but contrasted by pro-oxidant molecules. Cancer stem cells, which give rise to chemo/radio resistance and relapses, are characterized by a reducing environment, therefore being sensitive to the cytotoxic effects of oxidative damage [2]. Cancer tissue is characterized by a very low level of lipid peroxidation compared with the surrounding healthy tissues, as shown by biopsy analysis of 120 patients with primary hepatocarcinoma [3]. Accordingly, today the increase in oxidative damage represents a possible strategy for cancer therapy [4]. Intracellular generation of reactive oxygen species has been proposed as a possible Trojan horse for eliminating cancer cells [5].Several attempts have been made in order to use ozone as a source of oxidizing species in cancer therapy. Ozone was used as a gas [6] or as an ozonized aqueous solution [7]. These studies reported specific cytotoxic effects of ozone on cancer tissues without any damage in healthy tissues. However, the therapeutic effect obtained was meaningful but transitory. When used as a gas or aqueous solution, ozone, due its volatility, displays transitory effects only in the extracellular environment. Cells are surrounded by a lipophilic membrane hampering the entry of gas or water. Cancer cells are well equipped with antioxidants molecules to counteract cancer chemo/radio therapies. Accordingly, an oxidative intracellular environment may be effective in counteracting cancer chemo/radioresistance. Furthermore, the setting up of a high level of oxygen in the cancer mass may be useful for preventing metastases development. Indeed, low oxygen availability (hypoxia) is the main mechanism triggering the migration of cancer cells from the primary site. Hypoxia promotes cancer invasion and metastatisation by activating the met oncogene [8]. Hyperbaric ozone has been proposed as a possible tool for preventing these events [9].In this context, microRNA-based epigenetic regulation plays an important role. MicroRNA alterations drive cancer cells towards chemo/radioresistance [10] and modulate oxidative stress in cancer cells [11]. miR-146 plays a fundamental role in lung cancer progression and chemoresistance [12]. Increase in antioxidants such as reduced glutathione induces multi-drug resistance in neuroblastoma [13]. Downregulation of miR-15 and miR-16 is associated with increased availability of reduced glutathione in neuroblastoma cancer cells, contributing to chemoresistance [14].The antioxidant environment characterizing cancer cells is related to the constitutive overexpression of Nrf2; the pivotal transcription factor regulating the activation of antioxidant response elements [15]. Nrf2 activity provides growth advantage by increasing cancer chemoresistance and enhancing tumor cell growth [16]. Overexpression of Nrf2 in cancer cells protects them from the cytotoxic effects of anticancer therapies, resulting in chemo- and/or radioresistance [17].In this context, the herein presented study aimed at evaluating the efficacy of pharmacological preparations composed of ozonized oils in counteracting the growth of cancer cells. Ozonized oils are very stable molecules displaying antimicrobial properties [18]. The main goal of our study was the development of high-ozonide ozonized oil (HOO) to deliver a high amount of ozone-derived oxidizing species in a lipophilic complex able to penetrate the cancer cells and to activate apoptosis without damaging healthy tissues. Ozonized oils consist of unsaturated fatty acids that have been subjected to the action of ozone. The ozone is added to the double carbon–carbon bonds, with the formation of molozonides. These molecules quickly rearrange themselves according to the Criegee mechanism, causing the formation of trioxanes. Ozonides are generally unstable, while trioxanes are relatively stable but decompose under the action of reducing agents or intracellular enzymes. When the addition of ozone to the oil reaches the saturation of the double bonds, the viscosity increases with the progressive formation of ozonides until the oil reaches the consistency of gelatin. The peroxides contained in the oil can be hydrolyzed, giving rise to aldehydes and ketones with shorter chains compared with the original fatty acid. The length of the residues is determined by the position of the double bond along with the chain that reacted with ozone. The goals of the herein presented experimental study were the evaluation of HOO: (a) anticancer efficacy in vitro in cultured cancer cells; (b) molecular mechanism of action at the intracellular level; (c) mechanism of action at the systemic level; (d) inability to damage non-cancer cells; (e) anticancer efficacy and safety in vivo in human subjects and cancer patients.HOO was tested in human lung adenocarcinoma cells (A549 cell line) (IRCCS Policlinico San Martino, Genoa, Italy), and they were grown in D-MEM (GIBCO Invitrogen, Milano, Italy) containing 10% fetal bovine serum (Sigma-Aldrich, Milano, Italy) at 37 °C in 5% CO2 and 100% humidity. Non-ozonized sunflower oil was used as comparative sham control. The experiment was performed in quadruplicate. Glioblastoma U87MG cells were made available from the Biological Bank and Cell Factory of the IRCCS Ospedale Policlinico San Martino, Genoa, Italy. They were grown in DMEM high glucose media (Sigma-Aldrich, Milan, Italy), supplemented with 10% fetal calf serum (Euroclone, Milan, Italy), 2 mM L-glutamine (Euroclone, Milan, Italy), and 1% penicillin–streptomycin (Euroclone, Milan, Italy) at 37 °C in 5% CO2 incubator. Cells were seeded in 96-well plates at a density of 6 × 103 cell per well in 100 uL of culture medium, and they were treated with 10% HOO for 2, 6, 12, and 24 h. Sham-treated cells were used as control. After treatment, cells were washed with PBS (Euroclone, Milan, Italy), fixed and stained with a solution of crystal violet containing methanol 20% v/v. The following day, a solution of acetic acid 30% v/v in water was added, and samples were read by a microplate photometer (Multiskan FC, Thermo Scientific) at 570 nm.Cell viability was determined by Trypan blue staining (labeling dead cells) and MTT test (labeling viable cells), as previously reported [19]. The dynamics whereby the HOO formulation induces the killing of cancer cells were examined using a normal and trichrome fluorescence microscopy. The nucleus was stained in blue by DAPI (Sigma), the mitochondrial membranes in green by DiOC6 (Sigma), and calcium release into cytoplasm in red by Rhodamine2 (Sigma). HOO was stained by red Nile dye (Sigma) to trace its penetration inside cell cytoplasm.The HOO mechanism of killing cancer cells was also explored by a field emission scanning electron microscope (FE-SEM, Zeiss Supra 40VP, Carl Zeiss, Germany) equipped with energy dispersive X-ray analysis (EDX) microprobe for elemental analysis (Oxford “INCA Energie 450 × 3”, Oxford Instruments, UK), comparatively examining sham-treated and HOO-treated A549 cancer cells. EDX elemental analysis was performed at high magnifications (10,000×) with a spot at the middle of the cells before and after the HOO treatment.HOO was added to cell culture medium in order to discriminate the prevalent cell death mechanism between necrosis and apoptosis. A549 human lung adenocarcinoma cells were purchased from the Biological Bank and Cell Factory (IRCCS Policlinico San Martino, Genoa, Italy). They were grown in DMEM medium (Sigma-Aldrich, Milan, Italy), supplemented with 10% fetal calf serum (Euroclone, Milan, Italy), 2 mM L-glutamine (Euroclone, Milan, Italy), and 1% penicillin–streptomycin (Euroclone, Milan, Italy) at 37 °C in a 5% CO2 incubator. The day before the experiment, A549 cells were seeded in a 6-well plate at a density of 8 × 104 cells per well in 3 mL of culture medium DMEM (Sigma-Aldrich, Milan, Italy). After twenty-four hours of seeding, cells were treated with 10% v/v of HOO for 2 h and 4 h. Then, Muse™ Annexin V & Dead Cell Assay was performed. Cells were dissociated from each well to obtain single-cell suspensions, and 100 μL of these suspensions was added to each tube, together with 100 μL of the Muse™ Annexin V & Dead Cell Reagent (BD Biosciences Pharmingen 2350 Qume Drive San Jose, CA, USA). The samples were mixed thoroughly by vortexing and then stained at room temperature in dark for 20 min before being analyzed by flow cytometry (FACS Canto II cytometer, Becton Dickinson BD, Franklin Lakes, NJ, USA).Microscope examination of cell morphology showed that in cancer cells treated with ozonized oil, at 1 h cell viability is still maintained, while cell sufferance and lack of viability is massive at 24 h. Accordingly, the mechanisms causing loss of cell viability should occur in the 1–24 h time interval. Because the activation of apoptotic mechanisms requires at least 4 h, this was the timeline when we decided to evaluate this parameter.The 3D structure of cardiolipin, the main mitochondrion outer membrane monomer, was reconstructed. Cardiolipin structural variations were determined according to the presence or absence of cytochrome c binding in normal and cancer cells, respectively. The binding between cardiolipin and HOO under all these conditions was analyzed. Cardiolipin bilayer was built using the Membrane Builder generator from the CHARMM-GUI web toolkit [20,21]. A bilayer was built using a deprotonate cardiolipin molecule in the same way. We minimized both these structures by performing a short molecular dynamic (MD) simulation using the software GROMACS [22]. For both systems, the pre-production minimization and relaxation protocols were automatically generated by the CHARMM-GUI Input Generator. They consisted of 5000 steps of energy minimization, keeping a constant volume and a temperature of 303.15 K (NVT) using the Berendsen thermostat for 20,000 steps with a 1 fs time step. The production runs adopted a time step of 2 fs. In this case, each MD was launched for total 1 ns of simulation.The relationship between ozonide amount and cancer cell-killing effect was examined by analyzing comparatively 9 ozonized oils having different levels of ozonides. The following formulations were evaluated by comparing them to the untreated control or to control treated with non-ozonized sunflower oil (sham-control): ≤100 ozonides (Ozone Elite oil, Ozone cream oil 10, Oil olive O3 TuPiel), ≤300 ozonides (VO3 active spray, Prog. Olive oil, EMI sunflower oil, Oil Ozofarm), ≥700 ozonides (HOO 700), ≥1100 ozonides (HOO 1100).Anaplastic carcinoma cells A549 were grown in the presence of different ozonized oils, as previously reported; their ability to induce cell death was assessed by crystal-violet viability assay. For each formulation, the quantity of cells still vital after treatment (percentage as compared with the untreated control bearing 100% vitality) was evaluated. All the experiments were replicated 8 times for a total of 88 independent experimental analyses (11 experimental conditions × 8 replicates).The inability of HOO to induce cytopathic effects in healthy cells (safety) was tested in primary differentiated human keratinocytes (Biological Bank and Cell Factory, IRCCS Policlinico San Martino, Genoa, Italy) treated for 1–3 h with HOO, 80% v/v with the culture medium. The results were examined at 48 and 72 h after treatment.This experiment was performed to solve the problem of HOO application-timing in relation to radiotherapy, i.e., whether a synergistic effect in killing cancer cells exists or not. In the case of positive answer, it should be clarified whether to apply HOO before or after the treatment with gamma radiation. To face these problems, an in vitro experiment was performed in anaplastic carcinoma cells (A549) exposed to ionizing radiation (2 Gy) undergoing HOO treatment either before or after radiation treatment. Cell survival was evaluated by crystal violet staining, and results obtained in OHOO-treated cells compared with those obtained in control cells exposed to radiation and treated with sunflower seed oil (sham-control). The experiment consisted of 16 replicates in multi-well plate for a total of 48 experiments (3 experimental conditions × 16 replicates).The use of ozonized oil per os in human subjects as a food integrator was approved by the Health Ministry of Malta (approval number 0075/2020 according to EC1924/2006) issued on 17 March 2020.Five healthy male subjects aged 49.2 ± 12.7 years old were treated for 1 week, administering 12 mL of HOO per os per day. Blood samples were collected before administration (T0) and after treatment (T1) and used for immunological analyses by FACS. The influence of HOO on blood monocytes was performed using HLAdr monocyte activation marker. NK and helper lymphocytes counts were performed using CD3 and CD4 markers. FACS analyses were performed using a LSR Fortessa X20 (Becton and Dickinson, Eysins, Switzerland).A total of 115 cancer patients (age 58 ± 14 years; 44 males, 71 females) were treated with HOO contained in cellulose pills for 8 months as complementary therapy. The treatment was performed in parallel to standard chemo/radio therapeutic regimens performed for each cancer type according to the international guidelines.The cancer diagnoses were the followings: brain (glioblastoma and astrocytoma) 22, pancreas adenocarcinoma 18, skin epithelioma (squamous and basal) 7, lung (NSCLC and small cell lung cancer) 12, colon adenocarcinoma 13, breast cancer (estrogen receptor positive) 24, prostate adenocarcinoma 7 (Gleason severity score >8), ovary and womb 5, kidney and bladder 5, non-Hodgkin’s skin lymphoma 2.Cancer status at T0 (before HOO administration) and T1 (after HOO treatment) was examined by NMR, TAC, and PET, as performed for standard follow-up regimens according to international guidelines. Blood analyses were performed monthly. Amounts of oxidant (H2O2 milli-equivalent per 100 mL) and antioxidant (ascorbic acid milli-equivalent per 100 mL) in blood were examined monthly by the Free Radical Analysis System using a Fras4 Evolvo System (H&D, Parma, Italy) [23,24].A battery of oils having different ozonide content was tested for its ability to kill A549 cancer cells by evaluating the decrease in cell viability. The results obtained are shown in Figure 1.Cancer cell viability was observed to be decreased in the presence of ozonized oils compared with controls. However, ozonized oils at low ozonide were unable to reduce the percentage of surviving cells to under 10%. The only formulations able to achieve this result were the HOOs. In fact, the percentage of surviving cells was only 6.8% in HOO 700 ozonide and reached the minimum value detected of 2.7% in HOO 1100 ozonide. The EC50 was calculated according to the exponential regression equation between ozonide and cell viability, obtaining a value of 433 ozonides. It is noteworthy that the dose–response effect observed for HOO depends on the amount of ozonide. This experiment showed that the amount of ozonide is the key element of the killing effect of ozonized oils against cancer cells.Accordingly, HOO 1100 was selected for further analyses because of the highest level of cancer cell killing efficacy as compared with the other ozonized oils. Indeed, in HOO, ozonide content is equivalent to 800 meq O2/kg, corresponding to 220 mg of O3.Human lung cancer A549 untreated cells and cells treated with sunflower oil (sham) rapidly grew and reached to confluency after 72 h. Conversely, A549 cells treated with HOO showed impaired growth during first 24 h. After this time, they rapidly underwent cell death, which culminated at 72 h. Cell death was characterized by (a) disappearance of the cell-growth carpet; (b) presence of dead cells in the supernatant; (c) diffuse apoptotic bodies. These results are shown in Figure 2. The lack of viability of cancer cells treated with HOO was also demonstrated at 24 h by Trypan blue staining selectively labelling only death cells (Figure 3).After 12 h of HOO treatment, U87MG showed 4.68% of cell viability, as evaluated by MTT test, while A549 showed 8.43% of cell viability. After 24 h of treatment, U87MG showed 7.69% of cell viability and A549 showed 12.16% of cell viability. Accordingly, glioblastoma U87MG cells were more sensitive to HOO than lung adenocarcinoma A549 cells. This finding was well evident after 12 and 24 h of treatment. Regarding shorter time, there was no significant time difference between the two cell lines tested. Indeed, after 2 h the percentage of live cells was 13.91% for U87MG and 13.51% for A549; after 6 h of HOO treatment U87MG showed 6.23% of cell viability, while A549 showed 7.58%. These results are demonstrated in Figure 4.Sunflower seed oil and HOO penetration inside A549-treated cells was traced by microscope light scattering. Sunflower seed oil (sham-control) penetrated the cells (cytoplasm) only in a minimal amount and was compartmentalized (closed) into small well-defined vacuoles. Conversely, ozonized oil (HOO) penetrated abundantly into the cytoplasm, likely due to its peculiar ability to oxidize cell membranes, in particular the plasmatic membrane that delimits the cell from the external environment. Once penetrating the cytoplasm, the ozonized oil was initially compartmentalized into vacuoles; however, the membranes of these vacuoles were rapidly oxidized, and the oil spread into the cytoplasm overlapping intracellular membranes and organelles (Figure 5).These events culminated after 24 h and were followed by death of cells treated with HOO.The dose-dependent release of intra-cytoplasmic calcium was detected by rhodamine staining in cancer cells treated with HOO (Figure 6A). In the same cells, the mitochondrial membranes were selectively stained by green dye (DiOC6), verifying the decrease in signal intensity in HOO as compared with control cells treated with oil only (Figure 6B). This result reflected the damage of mitochondrial membranes, as induced by ozonized oil treatment. In cells receiving this treatment, it was also possible to observe the presence of large lipid vacuoles mainly located in the perinuclear zone (Figure 6, arrows). We carried out the characterization of these vacuoles by labeling them with LC3 in order to verify whether they were autophagic vacuoles, but the results were negative. Vacuole staining performed using dyes for lipid materials was not effective due to the rapid oxidative degradation of the dyes (Nile red) that was observed. Therefore, it is likely that these vacuoles are composed of oxidative lipids that cannot be catabolized by the cell. Because of this reason, the HOO-treated cancer cell takes the characteristic appearance of a ‘foamy cell’, which is characteristic of cells that accumulate oxidized lipids but cannot catabolize them due to the high level of mitochondrial damage. Indeed, lipid catabolism through the beta-oxidation biochemical pathway occurs inside mitochondria. The presence of these vacuoles in HOO-treated cells further demonstrates that the HOO action is expressed preferentially and directly towards the mitochondrial membranes.In Figure 6C, fluorescence in cultured cells was evaluated by opening both channels (red and green) in order to check the overlay between mitochondrial membrane damage (attenuated green light) and extramitochondrial calcium release (red). In the case of overlap, the resulting color was yellow, otherwise the green and red colors were maintained.In HOO-treated cancer cells, the images showed a yellow staining. This result indicates that the release of intracellular calcium (red) took place exactly from the mitochondrial membrane (green) damaged by HOO. In fact, this overlap (yellow color) did not exist in control cells where the mitochondrial membranes were not damaged, and there was no release of calcium from the mitochondria. Accordingly, the green color was maintained and no red color was shown. This result indicates that in cancer cells treated with sunflower oil (sham-control), calcium remained compartmentalized inside mitochondria; conversely, in HOO-treated cells, calcium was abundantly released from the mitochondria and spread into the cytoplasm. This result shows that the intracellular induction of mitochondrial damage with the consequent activation of intrinsic apoptosis was the main mechanism underlying the anticancer action of HOO.Figure 7 shows the FE-SEM images of both the control and HOO-treated cells. At low magnification (a, e), the change of morphology induced by the ozonide treatment on the cells is evident. The HOO-treated cells underwent dramatic smoothing and rounding (b, f), disintegration and death (c, g), and the size decreased (d, h).The number of elements contained in cells was estimated by EDX analysis. The carbon/oxygen ratio was high (C/O 4.3) in cancer cells, a situation envisaging the presence of the reducing intracellular environment characterizing these cells (left panel). Conversely, the carbon oxygen ratio became extremely low (C/O 1.5) in HOO-treated cancer cells, a situation envisaging the occurrence of oxidative stress in the intracellular environment as well as lipids and carbon-chain structure oxidation (right panel).Annexin V & Dead Cell Assay showed that apoptosis is the main mechanism of cell death affecting A549 cells treated with both HOO700 and 1100. Indeed, after two hours of treatment, control cells showed 11.10% of total apoptosis, with a clear prevalence of late apoptosis (40.90%) on early apoptosis (3.25%). Cells treated with HOO700 showed 30.35% of total apoptosis, divided into late apoptosis (29.10%) and early apoptosis (1.25%). HOO1100 determined 46.20% of total apoptosis (45% of late apoptosis and 1.20% of early apoptosis). These results are shown in the upper panels of Figure 8.Regarding 4 h of treatment, in control cells, 9.30% of total apoptosis (7.35% of late apoptosis and 1.95% of early apoptosis) was detected, 50.80% in cells treated with HOO700 (46.95% late and 3.85% early apoptosis) and 47.65% in cells treated with HOO 1100 (45% late apoptosis and 2.65% of early apoptosis). These results are shown in the bottom panels of Figure 8.The results of the molecular dynamic simulations suggested two different conformations for the processed systems. In the case of active mitochondrion (healthy cell), the cardiolipin bilayer was thick, tight, and symmetrical, with no breaks of continuity between the hydrophilic heads protruding into the cytoplasm. This solid structure prevented the access to the hydrophobic tails of cardiolipin of the oxidizing radicals present in cytoplasm, such as those carried by OHOO and indicated by red circles in Figure 5. Therefore, the mitochondrion of the normal cell was resistant to the killing effects of OHOO. This structural situation is reported in Figure 9 left panels.Conversely, in cancer cells, cardiolipin modified its structure due to the absence of the interaction with a functioning cytochrome c (the pivotal effector of aerobic glycolysis). Under these circumstances, the angle of convergence of the hydrophobic legs with the hydrophilic head was increased, a situation resulting in the divarication of the hydrophobic tails. This phenomenon occurred in all the cardiolipin monomers of the mitochondrial membrane, amplifying this molecular variation on the whole mitochondrial membrane. Thus, the mitochondrial membrane in cancer cell appeared at bioinformatic model thinner than in normal cells. This divarication of the hydrophobic tails created breaks of continuity between the hydrophilic heads of the cardiolipin that protrude into the cytoplasm, allowing access of HOO-oxidizing radicals to the hydrophobic tails of cardiolipin. According to this computational model, the mitochondrion of cancer cell was specifically sensitive to the damaging effects of HOO. This structural situation is reported in Figure 9 right panels.Results indicate that neither alteration of cell viability nor cytopathic effects occurred in noncancer cells treated with HOO, as demonstrated in skin keratinocytes (Figure 10).Cell viability of keratinocytes treated with ozonized oils quantified by MTT test was 100% in control cells, 99.8% after 1 h, 99.4% after 2 h, 98.7% after 3 h in ozonized oil-treated cells. Cell viability was not evaluated at times >3 h because the oil interface blocked cell exchange with culture medium, causing cell sufferance both in sham-treated cells (sunflower oil) and ozonized oil-treated cells, independent of oil toxicity.A strong radio-sensitizing effect of HOO 700 and even more of HOO 1100 was detected. The maximum effect observed was obtained treating A549 cancer cells with HOO 1100 after their exposure to gamma rays. From a mechanistic point of view, this effect was in line with the activation of the intrinsic mitochondrial apoptosis activated by HOO in cancer cells undergoing genotoxic damage induced by radiotherapy. The results obtained are shown in Figure 11.A summary of the results obtained in vitro is reported in Table 1.Two volunteer healthy subjects (males, 55 years old) were treated with 12 mL of HOO 700 for 1 week twice per day away from meals. Cytofluorimetric analysis (FACS) was performed to evaluate macrophages and lymphocytes pro-inflammatory activation in the peripheral blood before (T0) and after treatment (T1). The results obtained showed a marked decrease in macrophage activation markers (HLAdr) in both subjects, while no variations were observed in the lymphocyte subpopulations responsible for the protective immunity. An example of the cytofluorimetric results obtained is reported in Figure 12. Cytofluorimetric analysis evaluated T-helper CD4+ and cytotoxic T lymphocytes CD8+ before and after ozonized oil treatment without observing any variation. These results provide evidence that HOO can induce anti-inflammatory effect without causing immuno-suppression.Drug safety was also evaluated in vivo by analyzing the traditional blood chemistry parameters of the two volunteers treated: no alterations were found in the basal physiological state.The transferability of the results obtained in vitro and in vivo in animals was initially tested in seven human patients affected by cancer. The results obtained are summarized as follows.The patient was a 93-year-old female with malignant spino-cellular epidermoidal carcinoma, confirmed histopathologically. The neoplasia was localized in the scalp in the parietal region and presented ab initio a character of extreme invasiveness and rapid progression. In fact, in only a few weeks, the cranial case was invaded with consequent parietal osteolysis. The neoplasia continued its progression rapidly, penetrating the skull and taking on the arachnoid. These data were revealed by computerized axial tomography (TAC). The neoplasia had considerable size in the outer part, thus covering the whole skull with conspicuous growth—not only endophytic (inside the skull) but also exophytic (protrusion of the neoplastic mass outside the skull).The cancer was highly malignant, characterized by a high level of anaplasia, rapid progression, presence of neoplastic ulcers of significant size, high inflammation of the peri-lesioned margins, and total absence of repair by granulomatous tissue in the areas surrounding the neoplastic ulcer. The baseline clinical situation, as observed at the patient’s first visit, is shown in Figure 13.Therefore, it was decided to use HOO 1100 in a post-treatment regimen—i.e., at the end of each radiotherapy session. Thus, at the end of the first radiotherapy session, OHOO 1100, gelled at 4 °C, was applied to the neoplastic lesion through a glass depositor, and the ulcer was coated with gauze and hydrophobic bandage.The treatment continued for eight consecutive sessions with 2 days of interval between each one. OHOO medication was renewed at the end of each radiotherapy session. Therefore, OHOO was left to act in the pathological area for 48 h in the interval between radiotherapy sessions.At the second session, the presence of yellowish exudate already present before the apposition of HOO was observed.The treatment with radiotherapy and HOO 1100 continued for a total of eight sessions. Changes in the neoplastic lesion following treatment are shown in Figure 13.Microscopic examination showed a strong size decrease in the cancer mass both in amplitude and in depth. Furthermore, the formation of a granulation repair tissue at the margins of the lesion was observed.The radiotherapy was then suspended, and only topical treatment with HOO 1100 continued. Medication frequency was reduced to once per week because of the impossibility of the caregiver to bring the patient to the hospital more frequently. Despite the discontinuation of the radiation treatment, the cancer did not grow but further continued its regression, as shown in Figure 13.These last results demonstrate the specific inhibitory effect of HOO 1100 against the tumor, even in absence of the radiation treatment. At the end of the follow-up, patients showed only a soft eschar of exudative material resulting from the colliquative necrosis of the neoplastic tissue. The eschar was not cleaned because the disappearance of the neoplastic mass left exposed the subarachnoid arteries, which were pulsating below the eschar itself. This result showed that the disappearance of the tumor mass occurred not only at the esophitic but also at the endophytic level. This result is remarkable because only a preparation characterized by a high bioavailability can be able to reach even the deepest areas of the tumor mass through a simple topical application.In order to establish whether the topical application of OOAO had a systemic effect on the oxidative balance, we performed free radicals’ analysis in blood plasma before (T0) the beginning and end (T1) of HOO treatment using a Fras 4 system. At T0, the parameters were dramatically altered with a particularly low value of oxidizing species, i.e., 38 U Carr (normal range 250–280 U). In parallel, the antioxidants were strongly increased, with a value of 8318 U Cor (normal range 2200–2800 U). The antioxidant/oxidant balance (UCor/UCar ratio) was therefore 219 (normal range 7–10). These results show that the presence of a cancer mass characterized by large size implied the strong decrease in oxidation at a systemic level. At T1, the values were changed as follows: 201 U Carr, 7544 UCor, and the ratio UCor/UCar was 37. Therefore, an increase of 530% of the oxidative species was observed, a decrease of 13% of the antioxidants, a decrease of 583% of the ratio UCor/UCar. These values indicate that the regression of the neoplastic mass was related to the variation of the oxidative balance induced by treatment with OHOO.At the end of the treatment the patient was in good health and no longer suffering or feverish. During treatment, no subjective or objective side effects related to OHOO treatment were observed.There was an 86-year-old female patient with a recurrent skin ulcer in the palmar region of the right forearm. The lesion was subjected to biopsy and histopathological analysis, and then it was classified as ulcerated metatypical nodular basocellular carcinoma; the lesion extended in depth, infiltrating the papillary and reticular derma; the presence of surrounding chronic inflammatory infiltrates was observed. Before treatment, the lesion was macroscopically characterized, as reported in Figure 14.Therefore, the lesion was definitively removed surgically. During treatment, no subjective or objective side effects were observed.Subject was a 55-year-old male affected by prostate cancer, as confirmed by biopsy and histopathological examination. Staging and severity were very high, with this cancer being classified with a Gleason score 9 (max scale value 10). Cancer dimension: 3.5 cm. Imaging (ECT) detected a trend towards invasion of the prostate capsule, peri-prostatic adipose tissue, left seminal vesicle, and local lymph nodes. Relatively low PSA (4.5 ng/mL) was observed due to the high anaplastic behavior and the poor differentiation. Before surgical treatment, the patient was treated for 60 days with daily intra-rectal administration of ozonized oil (HOO 700) and oral administration of ozonized oil 12 mL twice per day.At surgery, the field macroscopically appeared clear and without any evident inflammation or defragmentation of the cancer mass. A similar situation usually did not occur in such a high-grade malignancy characterized by high inflammation, severe infiltration of surrounding structures, fast growth, and extreme fragility of the cancer mass. Accordingly, surgery removed the whole cancer, as well as surrounding tissues, seminal vesicles, and lymph nodes. Microscope histopathological analysis did not detect any sign of inflammation in the cancer parenchyma or surrounding tissues. The presence of tumor-associated macrophages was not detected at all, at variance with the typical aspect of this high-malignancy cancer (Figure 15).Only 1 local lymph node, out of the 10 examined, was barely affected by cancer invasion.Patient subsequently underwent anti-hormone therapy and radiotherapy as scheduled by the standard treatment protocol. In parallel, he continued ozonized oil treatment. After 6 months, no sign of relapse was detected.The analysis of oxidative status in blood plasma indicated that oxidative stress was low before ozonized oil treatment (T0) (210 U Car, normal range 250–280) as well as the antioxidant/oxidant ratio (9.9 Ucor/UCar). After treatment, oil treatment (T1) oxidative stress was increased (295 U Car, 7.9 Ucor/UCar ratio). No adverse effect related to the ozonized oil treatment was detected.There was a 76-year-old male, affected by prostate cancer (adenocarcinoma), as confirmed by biopsy and histopathological examination (September 2015). No local or distant metastasis was detected. NMR revealed high metabolic rate, fast cell proliferation, and blood vessel proliferation. Biopsy detected severe inflammation of cancer mass and surrounding tissue. Intermediate malignancy: Gleason score was 6, and cancer dimension was 3.0 cm.No results were observed in anti-hormonal therapy (November–December 2015). Standard radiotherapy regimen was of 40 sessions (January–March 2016). In October 2016, after radiotherapy, NMR detected cancer persistence with cell proliferation and remarkable blood vessel proliferation.From January 2017, HOO oral treatment (12.5 mL twice a day for 18 months) was started. In February 2018, TAC and radioimaging demonstrated complete cancer disappearance (Figure 16).No relapses were insofar detected (ongoing follow-up, 18 months).No adverse effect related to the ozonized oil treatment was detected.There was a 74-year-old male affected by prostate cancer (adenocarcinoma), as confirmed by biopsy and histopathological examination. No metastasis was detected. High malignancy grade (Gleason 8), PSA 9.1 ng/mL was observed. Treatment with ozonized oil (intra-rectal once per day) and ozonized oral oil (12.5 mL twice a day) for 40 days before therapeutic surgery was started. At surgery, despite the high cancer malignancy, no invasion of prostatic capsule, surrounding adipose tissue, seminal vesicle, or lymph node was detected; blood vessel proliferation was observed. Full eradication was observed with no relapses after follow-up of 8 months.Female subject was a 3-year-old glioblastoma patient, as detected by TAC/NMR and confirmed by biopsy. High malignant grade was observed. Patient underwent standard radio/chemotherapy regimen. Ozonized oil was administered (12.5 mL twice a day for 90 days). The clinical follow-up was compared with those of three other young patients in a similar clinical situation but devoid of ozonized oil treatment. All of these three patients underwent fast cancer progression, and one died. Conversely, the ozonized oil led to the arrest of cancer growth and a dramatic decrease in cancer dimension that after 3 months of treatment was only 35% of that initially detected. A similar finding was totally unexpected in such a fast growing cancer.Cerebrospinal fluid was collected, and oxidative status was analyzed. The sample was contaminated by red blood cells. For these reasons, the analysis of oxidative status was unreliable. Conversely, after several efforts, we were successful in analyzing the antioxidant status, whose values were 892 and 895 (replicate analyses on 15 uL × 2) U cor (umol/L ascorbic acid equivalent). The normal reference value was 2500, with a b max–min range of 1800, which was below the threshold. Accordingly, the observed value was extremely low (despite the red blood cell contamination releasing antioxidant). It could be concluded that the depletion of antioxidant due to the therapeutically induced oxidative stress (ozonized oil treatment) had been effective on the CSF of this patient. This value (893 U cor) could be assumed as a threshold to be reached to obtain therapeutic effects.It is conceivable that ozonized oil, due to its high lipophilicity, is able to cross the blood–brain barrier. Specific in vitro and in vivo tests are ongoing to further substantiate this issue.Female subject was a 38-year-old affected by brain glioblastoma in left hemisphere (diagnosis July 2014. 1st NMR, (T0)). First surgery was performed in September 2014. High malignancy (grade III) was observed. In May 2017, relapses (2nd NMR) were detected, and in June 2017, there was a second surgery. September 2017 radiotherapy (60 Gy) in parallel with the start of ozonized oil therapy (oral administration, 6 mL per day) was performed. No relapse (September 2018) was detected (3rd NMR, T1). Cancer presence at T0 as well as its clearance at T1 are reported in Figure 17.After these 7 patients, a total of 115 cancer patients, 76 males and 39 females, average age 60.1 ± 17.8 years, were treated with ozonized oil. An overview of the results obtained with regard to oxidative status in cancer patients treated with HOO is reported in Figure 18. For reference, oxidative status was also analyzed in parallel in 40 cancer-free subjects (22 males, 18 females, average age 58.0 ± 6.4 years).Standard follow-up exams included hematological analyses, blood analysis of cancer markers (e.g., Ca-19), nuclear magnetic resonance, computerized tomography with and without glucose tracer, echography. Clinical outcomes observed in HOO-treated cancer patients as compared with cancer patients undergoing standard therapeutic regimens is reported in Figure 19, referring to all cancers. Data on clinical outcomes after treatment for each cancer type are reported in Table 2. Clinical outcomes were significantly different (chi-square p value < 0.0001) in cancer patients receiving standard treatment only as compared with those additionally receiving HOO as integrative therapy, as evaluated by chi-square test (Figure 19).Obtained results provide evidence that HOO is a new and interesting strategy for prevention of cancer progression and relapses. Indeed, patients undergoing HOO integration, in addition to standard therapeutic regimens, showed increased survival and decreased rate of relapses. These clinical outcomes are justified by the mechanisms analyzed in vitro to shed light on the effects of HOO in cancer cells. Cancer cells are highly sensitive to HOO oxidative effects due to their mitochondrial status. The oxidation of mitochondrial membranes can re-activate apoptosis in cancer cells [2]. The selective effect of HOO in killing cancer cells only without damaging normal cells is due to the different mitochondrial situation. The mitochondrion is inactive in a cancer cell, both as far as concerns metabolic (aerobic glycolysis) and pro-apoptotic function, that in normal cells is activated by the release of cytochrome c and calcium from the mitochondrion into the cytoplasm. This situation critically differentiates cancer from healthy tissue, explaining why cancer cells cannot die because of apoptosis while normal cells can. The outer mitochondrial membrane is predominantly composed of phospholipids, among which the most relevant is cardiolipin. This molecule is characterized by the presence of a hydrophilic phosphorylated head and two hydrophobic tails. Cardiolipin is organized to form the typical phospholipidic double-layer. However, the shape of cardiolipin is profoundly modified in relation to its binding to cytochrome c, a fundamental component of oxidative phosphorylation [26]. Cardiolipin displays its physiological structure only when bound to functioning cytochrome c or, in other words, in the active mitochondrion characterizing normal cells but not in cancer cells. The sensitivity to HOO is even higher for cancer stem cells that are antioxidant addicted. These cells have been selected among cancer cell pools because of their high level of antioxidant-based detoxifying mechanisms, allowing them to counteract the therapeutic effects of chemo/radiotherapies. The scavenging of these antioxidants that is exerted by HOO is an effective tool for making cancer stem cells sensitive to chemo/radiotherapies and for overcoming their resistance. This situation explains the improved clinical efficacy of standard therapeutic regimens in patients receiving HOO. This result may be achieved only by using ozonized oils having very high ozonide content, extremely pure, devoid of antioxidants, and highly bioavailable.HOO exerts anticancer effects by activating various protective mechanisms including (a) scavenging of antioxidants from cancer cells; (b) re-activation of intrinsic apoptosis; (c) inhibition of the activation of tumor associated macrophages; (d) increase of oxygen tissue availability decreasing angiogenesis and metastasis.An additional mechanism could be the competition with the mitochondrial fat oxidation metabolic pathway providing energy availability in cancer cells [27]. HOO is a metabolite of this pathway, but its catabolization inside the mitochondrion results in oxidative stress triggering mitochondrial damage, intracellular calcium release, and apoptosis in cancer cells. Targeting the intrinsic apoptotic pathway has been recently proposed as a new strategy against cancer [28].Cancer is a systemic disease. The neoplastic mass is not able to develop autonomously in the absence of the trophic support provided by neo-angiogenesis and inflammation. The presence of a conspicuous infiltrate of inflammatory macrophage cells is a specific characteristic of malignant cancer. These macrophage cells are unable to counteract the neoplastic growth, supporting it by supplying oxygen, metabolites, and neo-vessels. These cells are referred as tumor-associated macrophages. The degree of inflammation is one of the most predictive and prognostic indexes of the unfavorable development of a neoplasm. Therefore, cancer should be counteracted not only at a topical level but also at the systemic level, controlling inflammation and oxidative status in the whole organism. In this regard, HOO could represent a significant step forward. HOO kills cancer by targeting cancer stem cells and activating apoptosis but also by exerting anti-inflammatory effects at the systemic level. Herein, presented results indicate that HOO induces in vivo anti-inflammatory effects without causing immuno-suppression. The mechanism underlying this situation is the inhibition of macrophage oxidative burst. Activated macrophages release oxygen-reactive species and inflammatory cytokines in the tissue environment in order to neutralize bacteria, if present. However, in the absence of bacteria, macrophage activation leads to an inflammatory response, which can assume pathogenic relevance by promoting cancer growth and progression. HOO inhibits macrophage oxidative burst through a negative feedback mechanism; indeed, the presence of an extra-cellular environment enriched with ozone-oxidizing species blocks the release of further oxidizing species from the macrophages, thus inhibiting their activation and the consequent inflammation.The trophic support to solid cancers is provided by neoangiogenesis. This process is activated by low oxygen availability in cancer tissue, triggering hypoxia inducible factor representing the main activator of vascular growth factor release and blood vessel formation [29]. Herein, presented results provide evidence that HOO effectively releases oxygen species inside cancer tissue, thus counteracting the hypoxic situation triggering neoangiogenesis.In conclusion, the experimental studies performed provide evidence of the efficacy of HOO treatment in killing cancer cells, thus integrating and potentiating the therapeutic effects of standard therapies. This conclusion is supported by the biological plausibility of the specific mechanisms activated by HOO in cancer cells, mainly including mitochondrial damage and activation of apoptosis. These effects are exerted by HOO due to its peculiar characteristics including: (a) oxidant effectiveness due to the high level of ozonide content; (b) dose customization by evaluating oxidant/antioxidant balance in blood plasma as well as cancer stage; (c) anti-inflammatory effects; (d) increase in oxygen tissue availability, counteracting cancer metastasis triggered by local hypoxia; (d) capacity of penetration inside cancer cells.These findings provide evidence that HOO is endowed with a potential therapeutic efficacy against cancer in the absence of detectable side effects. Due to its pharmacokinetic and pharmacodynamic peculiarities, HOO represents an innovation in the field of complementary cancer therapy worthy of further clinical studies. Our result provides evidence that oral administration of ozonized oils with high ozonide content is a novel strategy for the prevention of cancer relapses and chemo/radioresistance. This approach could be used in clinical practice to fulfill the lack of anticancer treatments occurring in intervals between chemo/radio therapeutic regimens.Conceptualization, A.I.; methodology, A.I., A.C.; software, C.R.; validation, L.B.; formal analysis, A.I., S.S.; investigation, A.I., L.B.; resources, A.C.; data curation, E.F., A.S., M.C. (Massimo Chiara); writing—original draft preparation, A.I., A.P.; writing—review and editing, Z.K., M.C. (Matteo Congiu); visualization, A.C.; supervision, S.S., C.C., G.B.; project administration, A.P. All authors have read and agreed to the published version of the manuscript.This research received no external funding.The study was conducted in accordance with the Declaration of Helsinki, and the study was approved by the Health Ministry of Malta (approval number 0075/2020 according to EC1924/2006) issued on 17 March 2020.Informed consent was obtained from all subjects involved in the study.All data are available upon request to the corresponding author.The authors declare no conflict of interest.Comparative evaluation of ozonized oil effect on cancer cell viability (MTT test). Only ozonized oils having an ozonide content >700 decrease cancer cell viability below 10% in 24 h.Time-dependent in vitro growth of A549 lung cancer cells either sham-treated with sunflower oil (upper row) or treated with ozonide oil >700 ozonides with a percentage v/v of ozonized oil related to sunflower of 97% (lower row). Sham-treated cells grow rapidly reaching semi-confluence at 24 h and full confluence at 48 h. Cancer cells treated with ozonide oil already display impaired growth during the first 24 h. After this time, they rapidly undergo time-related increasing cell death, culminating at 72 h.Detection of non-viable cells by Trypan blue staining after 24 h since seeding of A549 cancer cells, either sham-treated with sunflower seed oil (left panels) or treated with ozonide oil >700 ozonides (right panels). Upper panels, standard microscopy; lower panels microscopy after Trypan blue staining. Death cells are detected only in ozonide oil treatment.Viability comparison between human lung adenocarcinoma (A549, A) and glioblastoma (U87MG, B) cell lines treated with HOO for different times (2, 6, 12, and 24 h).HOO penetration inside A549 lung cancer cells, as traced by microscope light scattering.Intracellular calcium release as induced by ozonized oil in A549 lung cancer cells detected by rhodamine staining (red) and fluorescence microscopy. Mitochondrial membranes are labeled by DiOC6 staining (green). Red and green overlap reading both channels at the same time (right columns) indicates that calcium is released from mitochondria (yellow). Nuclei are colored by DAPI staining (blue).SEM images of the cells before (A–D) and after the ozonide oil (>700 ozonides) treatment (E–H) at a magnification of ×2000 (A,E) and ×10,000.Apoptosis profile (Muse™ Annexin V & Dead Cell Assay) for A549 cells treated with 10% v/v of OOAO 700 and OOAO 1100 and untreated cells (K). Profiles were determined 2 h (upper panels) and 4 h (bottom panels) after treatment. Each plot has 4 quadrant markers, reflecting the different cellular states: the upper left quadrant contains dead cells (necrosis), the upper right has late apoptotic/dead cells (cells that are positive both for Annexin V and for cell death marker 7-AAD, 7-Aminoactinomycin D), the lower left contains live cells, and the lower right early apoptotic cells (cells that are positive only for Annexin V).Bioinformatic analysis of mitochondrial membrane in normal (left panels) and cancer (right panels) cells. Mitochondrial membrane monomer cardiolipin undergoes a 10 A increase in distance of hydrophobic tails when not bound with cytochrome c. This situation occurs in cancer cells, resulting in mitochondrial membrane thinning. Accordingly, only in cancer cells can ozonized lipid (red circles) reach the hydrophobic legs of cardiolipin, thus causing mitochondrial membrane damage. This situation does not occur in normal cells, where ozonized lipid cannot reach this molecular target.Ozonized oil (bubble oil in the culture medium 1, 2, 3 h) does not induce alterations in normal human keratinocytes. No change in cell viability, intercellular adhesion, and substrate adherence occur in ozonized oil-treated cells (Oil) as compared with untreated control.Synergism between gamma ray radiation and ozonized oils (OOAO) treatments in killing lung cancer cells.FACS analysis of blood monocytes (cd38, vertical axis). Activated monocytes are detected by analyzing HLAdr (horizontal axis). The number of HLAdr-positive activated monocytes was decreased (arrow) after 1 week of HOO treatment (T1, right panel) as compared with those detected in the same subject before treatment beginning (T0, left panel).Evolution of a radioresistant skin basal carcinoma in a female 93-year-old patient treated with HOO.Evolution of an ulcerated nodular basocellular carcinoma in a female 86-year-old patient. The lesion extended in depth, infiltrating the papillary and reticular derma; the presence of surrounding chronic inflammatory infiltrates was observed before treatment (T0). HOO treatment induced cancer regression and disappearance of the surrounding inflammatory halo after 60 days.Prostate cancer histopathological analysis detecting poor signs of inflammation in the cancer parenchyma and surrounding tissues despite the high malignancy grade (Gleason 9) after HOO treatment. The presence of tumor-associated macrophages was not detected at all, at variance with the usual aspect of this cancer type (magnification 100×).Disappearance of radioresistant prostate carcinoma (T0) after 1 year of HOO treatment (T1), as detected by nuclear magnetic resonance in a 76-year-old male patient.Disappearance of radioresistant brain glioblastoma (T0) after 1 year of HOO treatment (T1), as detected by nuclear magnetic resonance in a 38-year-old female patient.Analysis of oxidative status in the blood plasma of 115 cancer patients. The level of antioxidant was higher in cancer patients as compared with controls (left panel, blue columns). HOO treatment in cancer patients decreased the high level of antioxidant (left panel, red columns) moving back their amount to the level of unaffected controls.Clinical outcomes observed in HOO-treated cancer patients (red columns) as compared with cancer patients undergoing standard therapeutic regimens (blue columns), referring to all cancers.Summary of results from in vitro experiments dealing the effect of ozonized oil at high ozonides (HOO) in cancer cells.Clinical outcomes in cancer patients treated with HOO (follow-up 4 years).* Cancer stage at recruitment T4-N3-M1.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01175.txt

@@ -0,0 +1 @@
+Moderate hyperthermia is a potent radiosensitizer and its efficacy has been proven in randomized clinical trials for specific tumor entities. In spite of this, hyperthermia still lacks general acceptance in the oncological community and implementation of hyperthermia in clinical practice is still low. Reimbursement is one key factor regarding the availability of hyperthermia for deep-seated tumors, with high variability in reimbursement between countries. We report the current reimbursement status and related pattern of care for the use of deep hyperthermia in Switzerland over a time period of 4.5 years. This analysis will provide the basis for the national standardization of deep hyperthermia treatment schedules and quality assurance guidelines, as well as for the expansion of deep hyperthermia indications in the future. This comprehensive insight into deep hyperthermia reimbursement and practice in Switzerland might also be of interest for other national hyperthermia societies.Background: Moderate hyperthermia is a potent and evidence-based radiosensitizer. Several indications are reimbursed for the combination of deep hyperthermia with radiotherapy (dHT+RT). We evaluated the current practice of dHT+RT in Switzerland. Methods: All indications presented to the national hyperthermia tumor board for dHT between January 2017 and June 2021 were evaluated and treatment schedules were analyzed using descriptive statistics. Results: Of 183 patients presented at the hyperthermia tumor board, 71.6% were accepted and 54.1% (99/183) finally received dHT. The most commonly reimbursed dHT indications were “local recurrence and compression” (20%), rectal (14.7%) and bladder (13.7%) cancer, respectively. For 25.3% of patients, an individual request for insurance cover was necessary. 47.4% of patients were treated with curative intent; 36.8% were in-house patients and 63.2% were referred from other hospitals. Conclusions: Approximately two thirds of patients were referred for dHT+RT from external hospitals, indicating a general demand for dHT in Switzerland. The patterns of care were diverse with respect to treatment indication. To the best of our knowledge, this study shows for the first time the pattern of care in a national cohort treated with dHT+RT. This insight will serve as the basis for a national strategy to evaluate and expand the evidence for dHT.Moderate-temperature (39–45 degree Celsius) regional hyperthermia (HT) is concurrently applied with radiotherapy (RT) or chemotherapy [1]. Adding HT to RT improves treatment outcomes such as local tumor control or overall survival in specific tumor entities with a negligible toxicity profile [2,3]. HT can be applied with superficial HT devices for superficial tumors (less than 4 cm depth below the skin) or with deep HT (dHT) devices for tumors located at depth (more than 4 cm from the skin). Several techniques and devices for the clinical application of dHT exist [1,4,5]. Although its effect has been proven in several tumor entities with positive phase III randomized trials and meta-analyses [3], there is no widespread use in Europe. Reasons are multifactorial and have been previously summarized by Van der Zee et al. [1] and Overgaard et al. [6], but are still ”hot”. Briefly, not only proving that the tumor region was adequately heated but also to heat and sustain a uniform temperature in the tumor region are challenging as the body attempts to maintain temperature homeostasis. Some earlier trials with dHT reported questionable results with worse outcomes with dHT, most probably caused by insufficient heating, missing quality assurance and an imbalance in the patient groups ([7] and discussion in [8]). This confusion resulted in a persistent loss of credibility in the oncological community [6,8,9].Another reason for the lack of widespread availability is that HT, and especially dHT, is relatively labor-intensive and needs trained staff [1,10]. Furthermore, the use of dHT as a radiosensitizer competes with concurrent chemotherapy. The advantages of chemotherapy include easy administration, a lesser requirement of technical experience and comprehensive availability. The prime example of this is cervical cancer ([11], discussion in [12]). A financial obstacle is the uncertain cost reimbursement of HT treatment in most countries, limiting HT practice to university centers [8,9] and withholding it from the broader target population. Therefore, despite good but aged evidence, only a few dHT indications were incorporated into international oncology treatment guidelines.HT has a long tradition in Switzerland, starting in 1980 with the first clinical application of superficial HT with RT at the Center for Radiation-Oncology Kantonsspital Aarau. In 1988, the first dHT treatment in combination with RT (dHT+RT) was performed there. Superficial HT was later rolled out to a second hospital in Switzerland and clinical applications, mainly for recurrent breast cancer, were maintained at this site. Thus, prior to 2017, there were only two centers applying HT based on ESHO guidelines [13,14,15,16] in Switzerland (Kantonsspital Aarau and Lindenhofspital Bern), with only the Kantonsspital Aarau applying dHT. During this time, for every HT treatment, an individual request to the patients’ health insurance for reimbursement was required. The national Swiss Hyperthermia Network (SHN) was founded to synchronize and coordinate HT research activities at the national level, guarantee treatment quality and improve the evidence base for HT. In 2016, the SHN submitted a proposal for the reimbursement of HT+RT for selected evidence-based indications to the Swiss Federal Office of Public Health for superficial HT and dHT. Subsequently, four indications for superficial HT and five indications for dHT were temporarily approved for reimbursement for a period of two years as from 2017 (Table 1). It was stipulated that every patient receiving HT had to be presented to and have the indication confirmed by the national SHN tumor board, which was constituted by HT experts to guarantee the high quality of treatment decisions [17,18,19,20]. For patients who were likely to benefit from dHT+RT without a listed reimbursed indication, a specific request for insurance cover was necessary.At the end of the 2 years, the SHN submitted an update of the current evidence for dHT to the Swiss Federal Office of Public Health. After reevaluation, dHT indications were expanded in 2019 with the indications of “local tumor recurrence and compression” and “painful bone metastasis”, making a total of seven reimbursed dHT indications. As of July 2021, the Swiss Federal Office of Public Health granted unrestricted coverage for the dHT indications of “cervical cancer” and “painful bone metastasis”. Reimbursement for the dHT indications “local tumor recurrence and compression” and “soft tissue sarcoma” has been temporarily prolonged, again for another 2-year time period. The indications for bladder, pancreatic and rectal cancer lost their reimbursement status (Table 1) [20].Regarding superficial HT, four indications (specific situations in breast and head and neck cancer, malignant melanoma and palliative indications with local tumor compression), were granted for two years and then without time restrictions [17,19]. However, superficial HT is not within the scope of the present analysis.To the best of our knowledge, this is the first analysis of an unselected, dHT patient cohort regarding treatment indications, patient and tumor characteristics and treatment schedules. We aimed to perform a pattern of care analysis to shed more light on dHT practice in Switzerland and build a basis for a national strategy to evaluate, consolidate and expand the evidence for dHT.All patients presented at the SHN tumor board between January 2017 and June 2021 for the evaluation of radiative dHT+RT based on ESHO guidelines [13,14] were collected in a database. In July 2021, the reimbursed dHT indications changed and, since the end of 2021, a second center in Switzerland has started to apply dHT. This time period included a patient cohort treated by a single dHT center with only one modification of reimbursed dHT indications.Data from tumor board protocols were independently extracted and crosschecked by two authors regarding reimbursed dHT indications, patient and tumor characteristics and information regarding referring hospitals. These data then were crosschecked and completed with dHT and RT treatment details by three other authors. In case of any discrepancy, a consensus was reached. This project was approved by the local ethics committee (EKNZ2021-01022, 1 July 2021).Possible candidates for dHT were presented at the weekly national SHN tumor board by their referring physicians. The individual indication for dHT was discussed with at least two radiation oncologists with clinical experience in moderate dHT, including also senior medical oncologists. Indications were approved if the patient exhibited no contraindications for dHT (e.g., metal implant, cardio-pulmonary insufficiency, etc.), if dHT was technically feasible (only treatable lesions in accessible tumor locations) and if there was no other more appropriate treatment option (i.e., RT alone, hormone therapy, chemotherapy or immunotherapy).From 2017 to 2021, Kantonsspital Aarau was the only institution providing radiative dHT+RT in accordance with ESHO guidelines [13,14] and therefore received referrals from centers throughout Switzerland. Not only the optimal treatment sequence of HT and RT but also the optimal time interval between RT and HT or vice versa is still a matter of debate. Multiple working mechanisms requiring different optimal temperature ranges contribute to the effectiveness of HT, as comprehensively presented in Oei et al. [40]. In the absence of robust clinical data, the decision on the therapeutic sequence of HT and RT is made individually by the respective center. Preclinical studies indicated that the time interval between RT and HT should be kept as short as possible [41] but clinical studies addressing the time interval are sparse [42,43,44,45]. In two retrospective clinical studies investigating the effect of the time interval on treatment outcomes in cervical cancer patients, one revealed a strong correlation of a short time interval between RT and dHT for a better clinical outcome [44], where the other study showed that a time interval up to 4 h has no effect [45]. These contradicting results initiated a comprehensive discussion that depicted the complexity of this topic [46,47,48]. However, with regard to the dHT standard operating procedure at the Kantonsspital Aarau, dHT is given before RT with a minimal time interval.dHT was performed with the BSD 2000 3D Hyperthermia Systems© (BSD Medical Corporation/Pyrexar, Salt Lake City, UT, USA) using either the SigmaEye© or Sigma 60© applicator, depending on the diameter of the abdomen or limb. The interval between two dHT treatments was at least 72 h. For pelvic dHT, thermometry probes were inserted in the bladder, the rectum, the vagina, the anal margin and superficially on both groins for continuous thermometry and thermal mapping where possible/necessary. Interstitial thermometry was not performed except for patients receiving interstitial brachytherapy. For all other patients, the hyperthermia treatment planning software Sigma Hyperplan© (M/s Dr. Sennewald Medizintechnik GmbH, Munich, Germany) was used to estimate suitable power and steering parameters to achieve the targeted tumor temperature of 41 °C. A dHT session starts with a warm-up heating phase. The following plateau phase had a duration of 60 min and started when (a) the targeted temperature in the tumor was reached (this option was only possible if the heated tumor was adjacent to an intraluminal thermometry probe), (b) the targeted power and steering parameters were reached or (c) latest after a 30 min warm-up heating phase, respectively. During treatment, vital functions were continuously monitored.The frequency of dHT was determined individually. Usually, dHT once per week was used for curative indications and dHT twice per week for palliative indications. As not every patient started RT on a Monday, a reliable subdivision of dHT once versus twice per week was not possible. For the purpose of this study, dHT frequency was therefore categorized as once or once to twice a week. For patients referred from other hospitals, the optimal RT schedule in combination with dHT was discussed at the SHN tumor board; however, the final responsibility for the RT schedule lay with the referring center. Whenever possible, patients were treated within or analogous to an existing treatment protocol. Some patients treated for bladder, rectal, anal and pancreatic cancer received a trimodal treatment with dHT+RT and concurrent chemotherapy. These patients were treated within [49,50,51] or analogous to a clinical trial [50,51,52,53,54]. Patients were divided into “in-house” and “referred” patients. Every patient originating from the Kantonsspital Aarau was considered “in-house”. Additionally, patients from other hospitals without RT facilities, which referred patients for RT to the Kantonsspital Aarau, were also considered “in-house”. Patients from other hospitals with RT facilities who were referred for dHT were classified as “referred patients”, independent of where they finally received the RT treatment. To depict the spatial policy of referrals, referring hospitals were further divided into intra-cantonal and extra-cantonal and the distance by road from the referring hospitals to the Kantonsspital Aarau was calculated. There were three options for the organization of the dHT+RT treatment: (1) the patient received both dHT+RT at the Kantonsspital Aarau, (2) the patient received RT at the day of the dHT session at the Kantonsspital Aarau and the remainder of the RT at the referring hospital or (3) the patient received dHT sessions only at Kantonsspital Aarau and all RT sessions at the referring hospital. The latter option was deemed suboptimal based on the standard operating procedure at the Kantonsspital Aarau, wherein dHT should be given before RT with a minimal time interval. If not possible, a latency of 90 min between HT and RT was deemed acceptable. For patients treated with protons at the Paul Scherrer Institute, only option 3 was possible; however, the distance by road was less than 30 km. For referred patients, option 2 was preferred due to the short latency between RT and dHT. During the COVID-19 pandemic, this option was omitted to avoid mixing in-house and external patients to decrease the risk of infection. The time interval between dHT and start of the following RT was measured in patients receiving both dHT and RT at the Kantonsspital Aarau and was defined as the time between switching power off on the dHT device and first beam-on of the RT. Time points were extracted from automatical treatment recordings and stated in minutes.Descriptive statistics were used to describe patient and tumor characteristics and treatment details, which were presented as mean with standard error, median with (interquartile) range or frequencies with percentages, depending on their distribution. Data were represented using Statistical Package R (released 2021, 10 August, Version 4.1.1) and the ggplot2 package, version 3.3.5. Due to the combination of the small sample size, many stratification levels and wide heterogeneity of treatment and patient characteristics, statistical inference was not performed beyond the summary tables presented here as it was judged that a qualitative assessment of the data would be more suited to the aims of this study. Continuous values were summarized with mean, standard deviation, median and max/min values. Categorical variables were summarized as frequencies and proportions.The river plot was generated using the free, internet-based software SankeyMATIC [55].Between January 2017 and June 2021, 567 patients were presented for the evaluation of superficial or deep hyperthermia, with 32.3% (183/567) qualifying for dHT. Of these 183 patients, 28.4% (52/183) were deemed unsuitable. The remaining 131 patients were further assessed at a medical consultation and by their ability to tolerate the patient positioning required for dHT. This resulted in the further exclusion of 24.4% of patients (32/131). The reasons are stated in Figure 1a. In total, 54.1% (99/183) of patients initially presented at the SHN tumor board actually received dHT. Four patients had to be excluded due to withdrawal of consent, resulting in a total of 95 patients for analysis. Patients for superficial HT were beyond the scope of this analysis.The median age of patients receiving dHT was 65 years (range, 18–88). Moreover, 57.9% (55/95) of patients were male and 49.5% (47/95); 41.1% (39/95) and 9.5% (9/95) had an Eastern Cooperative Oncology Group (ECOG) performance score of 0, 1 or 2, respectively. A total of 47.4% (45/95) of patients received dHT with curative intent. Meanwhile, 42.1% (40/95) of patients had been previously irradiated and received dHT combined with re-irradiation (re-RT). In addition, 7.4% (7/95), 23.2% (22/95) and 69.5% (66/95) of patients were treated within a study protocol [49,50,51], analogous to a protocol [50,51,52,53,54] or as part of routine clinical practice, respectively (Table 2).Patients were divided into groups based on treatment indication regarding reimbursement status (reimbursed dHT indications vs. indication requiring an individual “request for insurance cover”) and based on primary tumor entities, respectively (Table 2, Figure 2, Supplementary Data, Figure S1). This revealed that “local tumor recurrence with compression” was the most common reimbursed dHT indication treated, representing 20.0% (19/95) of patients, followed by “rectal cancer” with 14.7% (14/95) and “bladder cancer” with 13.7% (13/95) of patients. Over the 4.5-year time period, 24.2% of patients (24/95) were treated with an indication not directly covered or not yet covered and therefore required an individual “request for insurance cover” to obtain reimbursement. Details of this patient group are provided in the Supplementary Data in Table S1. 15 of 24 patients who were treated from 2017 to 2018 and therefore before the two new dHT indications (“tumor local recurrence and compression” and “painful bone metastasis”) were added, as well as 9/24 patients in the time period from 2019 to the first semester of 2021. Ten of these 15 patients would have fallen within the two new indications, showing that the two new indications covered an existing demand. Regarding primary cancer entities, the most common was rectal cancer, with 22.1% (21/95), followed by bladder cancer with 15.8% (15/95) and soft tissue sarcoma with 13.7% (13/95) of patients (Table 2). Tumor entities with less than three treated patients are not individually represented but summarized in the group “others”, which contributed with 18.9% (18/95). Primary cancer entities, i.e., anal, colon and prostate cancer, presented in a clinical situation belonging to the reimbursed indications “local tumor recurrence and compression”, “painful bone metastasis” or to the group “request for insurance cover”. The time trend is shown in the Supplementary Data, in Figure S1.The patient population treated with dHT consisted of 36.8% (35/95) in-house and 63.2% (60/95) of patients referred from external radiation oncology institutions. To depict the spatial policy of referrals, the distance from the referring hospital to the Kantonsspital Aarau was calculated, resulting in a mean of 61.5 km (SD 54.3 km) and a median of 42 km (range 23–238 km) (Table 2).All in-house patients received their RT at the Kantonsspital Aarau. Regarding the patients referred from other hospitals, 23.3% (14/60) of them received both, dHT with all irradiations, at the Kantonsspital Aarau. Moreover, 10.0% (6/60) of patients received all irradiations at their referring hospital except at the day of dHT, where RT was applied at the Kantonsspital Aarau to minimize the time delay between HT and RT. In addition, 66.7% (40/60) of patients received only dHT treatment at the Kantonsspital Aarau and were irradiated at their referring hospital (Figure 3).Patient characteristics are described more in detail in Supplementary Table S2, comparing (1) in-house vs. referred patients, (2) patients receiving dHT in the setting of a re-RT vs. primary RT, (3) patients treated with palliative vs. curative intention or (4) patients treated within a clinical trial, analogous to a trial or in clinical routine practice, respectively (Supplementary Table S3A). Interestingly, (5) a gender difference was noted (Supplementary Table S4). One of the 95 treated patients stopped dHT+RT after three RT fractions due to reasons unrelated to treatment. This patient was excluded from treatment schedule analysis. In the whole cohort, a mean of 5.24 (SD ± 1.94) and a median of 5 (range 1–10) dHT sessions were applied, with 52.1% (49/94) of patients receiving it once a week and 47.9% (45/94) once to twice a week. Concurrent dHT was applied with external body RT (EBRT), stereotactic body RT (SBRT), protons and interstitial HDR-brachytherapy in 84% (79/94), 2.1% (2/94), 9.6% (9/94) and 4.3% (4/94) of patients, respectively. The mean total number of fractions was 21.7 (SD ± 8.89), with a median of 25 (range 4–38), a mean dose per fraction of 2.49 Gy (SD ± 1.35) and a median of 2 Gy (range 1.8–9 Gy). The mean total dose was 46.2 Gy (SD ± 12.8), with a median of 50 Gy (range 12.5–76 Gy). Moreover, 20.2% (19/94) of patients received an RT boost. RT was delivered daily in 83% (78/94) of patients (Table 3, Supplementary Table S3B). In total, 55 of 95 patients (57.9%) received dHT followed by RT at the Kantonsspital Aarau. The remaining 40 patients travelled to their referring hospital after the dHT session for the same-day RT (Figure 3). In the first group, the time interval between dHT and RT was available in 98.1% of patients (54/55). The mean and median time between the end of the dHT session and start of the RT was 19 min (SD ± 5.5) and 18 min (range 11–32 min), respectively. Evaluation of the time interval of the 40 patients receiving all RT at their referring institution was not possible due to the retrospective nature of this study and because these patients were irradiated at several RT facilities located all over the country. Treatment characteristics were compared between specific patient subgroups, including in-house vs. referred patients, primary RT vs. re-RT and curative vs. palliative intention (Table 3). The treatment schedules employed are stated per dHT indication and per individual patient in detail in Supplementary Table S5.The specific treatment schedules were dependent on the treatment indication, aim of treatment, pre-irradiation status, primary tumor entity and tumor stage. Patients treated with curative intent generally received a higher total dose, more RT fractions, usually 2 Gy per fraction and one dHT session per week. Palliative or re-RT treatment schedules mostly consisted of lower total doses, less RT fractions using moderate hypofractionation with 1–2 dHT sessions per week, but nearly the same total number of dHT sessions as in the curative setting. This coincides with the expected current practice in radiation oncology.The adherence to dHT was high, with 94% (89/95) of patients finishing all dHT sessions as initially prescribed. Six patients did not complete the prescribed sessions. Three of these six patients were treated for bladder cancer, two of them with tetramodal treatment (transurethral resection of bladder tumor (TUR-BT), chemotherapy, dHT+RT) and one with dHT+RT only. The reason for early discontinuation in these three patients was bladder irritation and/or bacterial cystitis, which prevented further catheterization for thermometry. Furthermore, 2/6 patients were treated for rectal cancer with local tumor recurrence with compression with palliative intent and were of ECOG 2. The reason for early discontinuation of dHT was deterioration of health status. The sixth patient was scheduled to receive neoadjuvant dHT+RT for soft tissue sarcoma of the limb. dHT was discontinued after the first HT session due to heat-induced pain in the tumor. During the investigated time period, only one RT center in Switzerland provided radiative dHT and seven dHT indications were approved for reimbursement in Switzerland. For other tumor situations that were likely to benefit from combined dHT+RT, an individual request to the patient’s insurance company was necessary. A prerequisite for coverage of the costs stipulated by the Swiss Federal Office of Public Health was the presentation and confirmation of the dHT indication at the SHN tumor board.Our analysis of the patient flow through this tumor board revealed a high number (approximately 50%) of patients who were not approved for dHT. This might be explained not only by the critical evaluation of the dHT indication by an expert panel, thus reflecting the quality of the tumor board decisions, but also by the fact that some referring physicians were not yet familiar with dHT as they presented patients with obvious contraindications, such as metal implants in the tumor region. We noted that only for two patients dHT could not be applied due to lack of cost recovery (Figure 1a), showing that health insurance companies in Switzerland will cover dHT when no other local treatment options than dHT+RT exist and the indication can be justified. The strict supervision of meaningful indications by the SHN tumor board probably contributed to the high acceptance rate of the health insurers. Therefore, we conclude that the SHN tumor board serves not only for the preselection of patients, besides contributing to the transparency and harmonization of treatment schedules, but also plays a role in teaching newcomers to the field. This analysis presents compelling evidence of an existing clinical demand for dHT for both palliative and curative indications. The majority (74.7%, 71/95) of patients in this analysis were treated based on the seven “reimbursed dHT indications” and only 25.3% (24/95) of patients required an individual “request to the insurance company” to cover the costs of therapy (Table 2). A closer look at the latter group revealed that, in the two years (2017 to 2018) before the introduction of the two new reimbursed dHT indications (local tumor compression and painful bone metastasis), more requests for dHT were submitted to insurance companies (15 vs. 9 patients). From 2017 to 2018, dHT was mainly prescribed for the two indications mentioned above (10 of 15) (Supplementary Table S1). With the approval of these two indications, the number of requests to insurance companies decreased, reflecting that an existing clinical demand had been covered. The linear time trend observed over the first two years, with an increase of one patient per semester, could be interpreted as epidemiological growth or may be due to the fact that hyperthermia achieved more visibility within the Swiss (radiation) oncology society. However, the COVID-19 pandemic has clearly influenced case numbers and indications treated from the first semester of 2020 onwards (Figure 2). Due to this confounding bias, a reliable time trend analysis of patient numbers was not possible; however, it is important to note that an uncontrolled increase in case numbers did not happen despite reimbursement of new treatment indications. Taken together, the dHT indications negotiated jointly by the Swiss Federal Office of Public Health and the SHN appear not to have induced a commercially driven increase in patients treated. With regard to the referral pattern, our analysis revealed that only 36.8% (35/95) of patients originated in-house and that 63.2% (60/95) patients were referred from external radiation oncology institutions (Figure 3). This shows that a dHT unit in Switzerland, even when integrated into a radiation oncology center, not only treats in-house patients. Patients have been referred for dHT from university hospitals and as well from the proton therapy center at the Paul Scherer Institute explicitly for the treatment of challenging oncological situations (Supplementary Table S2). This indicates that a dHT unit covers an existing demand for specific oncological situations, such as re-irradiation, organ-preserving treatment combinations (bladder and rectal cancer, soft tissue sarcoma) and other complex situations such as inoperable pancreatic cancer, soft tissue sarcoma or bulky, radioresistant tumors. In Switzerland, HT is frequently and incorrectly regarded as a mainly palliative treatment option. In the present analysis, we refute this by showing that 47.4% (45/95) of patients were treated with a curative treatment approach. The characteristics of the in-house patients revealed that they generally had a lower performance status and were more likely to be treated with palliative intent. Accordingly, dHT was more often used for the indication “local recurrence and compression”. Patients of low performance status are not fit to travel long distances for dHT, even if they would benefit from a radiosensitizer such as dHT, with its good toxicity profile. For palliative indications, the use of dHT could allow for a reduction in RT dose and thereby improve the tolerability and effect of RT, i.e., regarding pain relief, as has been shown by Chi et al. [39] for painful bone metastases. The referred patients in the present cohort travelled a relatively long mean distance of 62.2 km (SD ± 54.6 km), with a maximum of 238 km, to receive dHT (Supplementary Table S2). This effort is unreasonable for palliative and frail patients, which supports the future higher spatial availability of dHT units in Switzerland. The three most commonly reimbursed dHT indications were “local tumor recurrence with compression” (20%), “rectal cancer” (14.7%) and “bladder cancer” (13.7%) (Table 2). Unfortunately, the approval for reimbursement for the most common curative and organ-preserving indications, “rectal cancer” and “bladder cancer”, was withdrawn by July 2021 [20]. Patients treated for the dHT indication “rectal cancer” were mostly referred from external radiotherapy centers (Supplementary Table S2) and predominantly for re-irradiation (71.4%; 10/14 patients, data not shown). More than half (8/14 patients) were treated analogously to the HyRec trial [31] (Supplementary Table S5). The indication “bladder cancer” closes a gap in treatment options for either elderly and frail patients or patients seeking a bladder-sparing treatment approach. Patients were referred from external hospitals for these indications, underlining the demand for this treatment option as well. The SHN board is convinced that there is good evidence for dHT for these two indications [26,27,28,29,32], especially in rectal cancer, since two recent studies showed a promising effect of dHT [30,31]. Regarding the other dHT indications, the present analysis revealed that only a few patients are treated for the dHT indication “cervical cancer”, although it is associated with the strongest clinical evidence [21,22,23]. This could be explained by the low incidence of cervical cancer in Switzerland and the fact that this indication only receives direct reimbursement in the case of re-RT and for patients with contraindication to concurrent chemotherapy, which is rarely the case in Switzerland. This is in contrast to, for example, the Netherlands, where dHT is reimbursed in the primary treatment setting in combination with RT and brachytherapy based on evidence from randomized trials [11]. Another observation is the low patient numbers treated for “painful bone metastases”, although its superior effect regarding pain control was shown in a phase III randomized trial [39]. At Kantonsspital Aarau, the combination of dHT+RT for the indication of painful bone metastases was intended to be increasingly used in the future, because, with the longer survival of metastatic patients, long-lasting pain control is also becoming more important. However, because, during the COVID-19 pandemic, non-mandatory treatments were minimized and painful bone metastases could be often sufficiently treated with hypofractionated RT schedules alone, dHT was not offered. After returning to normality in the first semester of 2021, dHT patient numbers almost doubled (Figure 2), reaching the limited capacity of treatment slots for dHT. Therefore, patients with curative treatment indications were prioritized and dHT+RT again was not actively offered to patients qualifying for painful bone metastases. With the increasing dHT treatment capacity and controlled establishment of more dHT units in Switzerland, more patients with painful bone metastases could benefit from the increased analgetic effect of dHT+RT.The present patterns-of-care analysis was conducted as an inventory/survey of current practice and as the basis for a national objective to define standardized treatment schedules in Switzerland. All reimbursed indications, except for the indications “tumor recurrence and compression” and “painful bone metastasis”, showed relatively standardized treatment schedules in analogy to clinical trials (Supplementary Table S5). In contrast, the indication group “local tumor recurrence with compression” represents a patient collective with enormous heterogeneity regarding primary cancer entities, re-RT status, RT modalities and treatment schedules. The only common denominator is that they were treated mostly with palliative intent (Supplementary Tables S2 and S5). Importantly, these patients often have no other treatment option apart from dHT+RT and local treatment effect has a high impact on their quality of life. Withholding dHT+RT as a last treatment option from these patients would, in our view, be unethical. Because these patients frequently required individually tailored treatment schedules based on their previous treatment, the standardization of the treatment schedules, especially for clinical trials, would also be difficult. It is therefore clear that an analysis of dHT efficacy in this patient group is a challenge. A good example for the standardization of dHT+RT treatment schedules in patients with tumor recurrences is the subgroup of the HyRec trial from Ott et al. [31,52] and the schedule with 5 × 4 Gy once weekly combined with weekly wIRA superficial HT in recurrent breast cancer from Notter et al. [56] for superficial HT. Such innovative study designs and further treatment schedules are required to evaluate and consolidate the effect of dHT in these heterogeneous patient groups.To the best of our knowledge, we report the first retrospective analysis of an unselected national patient cohort treated with dHT, evaluating patient numbers over 4.5 years, specific treatment indications, patient characteristics, tumor entities, the referral practice and corresponding treatment schedules in Switzerland.Nearly 50% of patients were treated with curative intent. Around two thirds of patients were referred from external institutions from all over Switzerland, including from university hospitals and the proton therapy center, for challenging oncologic situations such as re-RT, complex palliative situations, organ-preserving treatment combinations (bladder and rectal cancer, soft tissue sarcoma) and inoperable, bulky or radioresistant tumors. This observation refutes the common prejudice, at least in Switzerland, that HT is only used for palliative situations and clearly underlines the medical need for the combination of dHT+RT.Patients treated within the reimbursed dHT indications with predominantly curative intent were homogenous subgroups with relatively standardized treatment schedules according to published clinical trials. On the other hand, the present patterns-of-care analysis revealed that patients treated within the two palliative reimbursed indications “tumor local recurrence and compression” and “painful bone metastasis” exhibit immense heterogeneity regarding patient characteristics and treatment schedules, demonstrating the need for standardization as a basis for future clinical studies.This analysis will provide the basis for standardized national dHT treatment schedules and quality assurance guidelines to consolidate and expand dHT evidence. We think that this insight into dHT practice in Switzerland could be of interest for centers interested in the implementation of a dHT unit and for other HT societies, especially regarding reimbursement policy, and could also foster international study collaborations.The following are available online at https://www.mdpi.com/article/10.3390/cancers14051175/s1, Figure S1: Number of cases presented at the Swiss Hyperthermia Network tumor board by primary cancer entity. Table S1: Patients and tumor characteristics of patients treated with deep hyperthermia who required an individual request for insurance cover. Table S2: Patient characteristics regarding referral status, re-irradiation status and by treatment indication. Table S3: Patient, tumor and treatment characteristics according to treatment protocol. Table S4: Patient characteristics by gender. Table S5: Deep hyperthermia and combined radiotherapy treatment schedules by specific reimbursed dHT indications.Conceptualization, E.S., E.P., S.B. and O.R.; methodology, E.S., A.K., S.B. and O.R.; software, A.K. and E.S.; validation, E.S., E.P., A.A., A.K., R.K., O.T., D.M., M.N., S.B. and O.R.; formal analysis, E.S., A.K. and O.R.; data curation, E.S., A.A., E.P. and R.K.; writing—original draft preparation, E.S.; writing—review and editing, E.S., E.P., A.A., A.K., R.K., O.T., D.M., M.N., S.R., S.B. and O.R.; visualization, E.S. and A.K.; supervision, S.B. and O.R.; funding acquisition, E.S. and O.R. All authors have read and agreed to the published version of the manuscript.This research was funded by a Scientific Association of Swiss Radiation Oncology (SASRO) research grant (to E.S.) and by the Swiss Hyperthermia Network (SHN). In addition, this research has received support from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie (MSCA-ITN) grant “Hyperboost” project, no. 955625 (to O.R. and S.B.).The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Ethics Committee Nordwest—und Zentralschweiz of Switzerland (protocol code 2021-01022, 1 July 2021).Informed consent was obtained from subjects involved in the study. Patients declining use of their data were excluded from analysis, as stated in Figure 1a.The data presented in this study are available on request from the corresponding author. The data are not publicly available due to privacy and ethical reasons.We thank Sonja Schwenne, for the administrative support.The authors declare no conflict of interest.Patient flow through the SHN tumor board. (a) Patients presented for dHT were excluded if dHT was not indicated (green) or a physical examination and treatment tolerability check revealed an exclusion criterion (yellow). Only patients with informed consent were eligible for analysis (violet). Background colors match the corresponding bar chart plot. (b) Patients presented at the SHN tumor board from January 2017 to June 2021 were depicted per semester. Events that may have affected the number of patients and indications treated were the two new “reimbursed dHT indications” as of 2019 and the changes in oncological treatment patterns during the COVID-19 pandemic, especially the COVID-19 lockdown in Switzerland (11 March to 26 April 2020; 1st semester 2020). Abbreviations: CI: contraindication, CIED: cardiac implantable electronic device, Claustroph: claustrophobia, dHT: deep hyperthermia, Sem: semester, SHN: Swiss Hyperthermia Network, Pts: patients, S1: 1st semester, S2: 2nd semester.Trend of patients treated with combined deep hyperthermia (dHT) and radiotherapy over time. Bar chart where numbers of patients receiving dHT between January 2017 and June 2021 are depicted per semester (S1 and S2) and divided into “reimbursed dHT indications” with specific subgroups and “request for insurance cover”. From 2017 to 2018, a linear increase in patient numbers with approx. 1 patient per semester was showed. Two new reimbursed indications, “local tumor recurrence with compression” and “painful bone metastasis”, were granted as from 2019 (blue shaded background). COVID-19 lockdown in Switzerland was during 1st semester 2020 (11 March to 26 April 2020).River plot showing the proportions of in-house and referred patients and where the RT and dHT were applied. On the left, patients are grouped according to source of referral. On the right, the three options regarding where and how dHT+RT treatment was applied are stated. The thickness of the connecting flowlines represents the proportion of patients. Abbreviations: dHT: deep hyperthermia, dHT+RT: combined dHT and RT, KSA: Kantonsspital Aarau (dHT center), RT: radiation.Indications for deep hyperthermia (dHT) with granted reimbursement in Switzerland [18,19,20] are stated with specifications and underlying evidence.Prior irradiationContraindication for ChTFunction preservationPrior irradiationContraindication for ChTFunction preservationLocal recurrence in pre-irradiated areaContraindication for ChTFunction preservationContraindication for ChTLocally advanced, initially inoperable tumorPatients with local tumor recurrence and symptoms due to tumor compression (palliative situation)Tumor depth > 5 cmLocated in the pelvis or vertebral bodiesTumor depth > 5 cmPrerequisites are (i) combination with radiotherapy (RT), (ii) the indication has to be presented and confirmed at the Swiss Hyperthermia Network (SHN) tumor board, (iii) the combined dHT + RT has to be performed at an institution affiliated with the SHN. The reimbursement status is indicated per time period and coded with underlying colors. Green = time-unrestricted reimbursement; yellow = reimbursed indications limited for two further years; red = indications no longer reimbursed; grey = initially not reimbursed indications (request for insurance cover was required). Abbreviations: ChT: chemotherapy, HT: hyperthermia.Patient and tumor characteristics with treatment indications, referral status and deep hyperthermia treatment adherence. Specifications of “reimbursed dHT indications” are given in Table 1.Abbreviations: dHT: deep hyperthermia, dHT+RT: combined dHT and RT, ECOG: Eastern Cooperative Oncology Group, intra and extra-cantonal: cantons in Switzerland are equivalent to states, provinces or regions in other countries, KSA: Kantonsspital Aarau (=dHT center), Others: the definition is given in the text, RT: radiotherapy, SD: standard deviation.Treatment characteristics for specific patient subgroups comparing in-house vs. referred patients, primary RT vs. re-RT and curative vs. palliative intention.One patient stopped treatment very early and was excluded from the treatment characteristics table. Abbreviations: dHT: deep hyperthermia, EBRT: external body radiotherapy, Gy: Gray, HDR: high dose rate, RT: radiotherapy, SBRT: stereotactic body radiotherapy, SD: standard deviation.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01176.txt

@@ -0,0 +1 @@
+Glioblastoma multiforme (GBM) is the most aggressive tumor type in the central nervous system. Hypoxia, defined as a lack of sufficient oxygen in tissues, is the most detrimental factor for the survival of GBM patients, promoting drug resistance, and invasion and inhibition of immune responses. Traditionally, tumor hypoxia has been studied from a narrow viewpoint, excluding the immune system and focusing primarily on the effect of hypoxia on blood vessels and tumor cells. More recently, however, evidence highlighting the important role of immunosurveillance has been uncovered for multiple tumors, including GBM. Thus, connecting the knowledge gained from traditional hypoxia studies with findings from recent immunological studies is urgently needed to better understand the role of hypoxia in cancer.Hypoxia is a hallmark of glioblastoma multiforme (GBM), the most aggressive cancer of the central nervous system, and is associated with multiple aspects of tumor pathogenesis. For example, hypoxia induces resistance to conventional cancer therapies and inhibits antitumor immune responses. Thus, targeting hypoxia is an attractive strategy for GBM therapy. However, traditional studies on hypoxia have largely excluded the immune system. Recently, the critical role of the immune system in the defense against multiple tumors has become apparent, leading to the development of effective immunotherapies targeting numerous cancer types. Critically, however, GBM is classified as a “cold tumor” due to poor immune responses. Thus, to improve GBM responsiveness against immunotherapies, an improved understanding of both immune function in GBM and the role of hypoxia in mediating immune responses within the GBM microenvironment is needed. In this review, we discuss the role of hypoxia in GBM from a clinical, pathological, and immunological perspective.Tumor cells have distinct metabolic features compared to normal cells. For example, although normal cells usually suppress glycolysis under normoxic conditions (i.e., the Pasteur effect), tumor cells preferentially use glycolysis despite the presence of oxygen, a phenomenon known as the Warburg effect [1]. Lactic acid accumulation resulting from the Warburg effect is metabolic hallmark of the tumor microenvironment (TME), leading to low pH. Critically, these unique metabolic characteristics can inhibit antitumor immune responses, making the TME more favorable for tumor progression [2]. Thus, a precise understanding of metabolic programming within the TME is essential for the development of effective antitumor therapy.Oxygen is the most basic and important component of cellular metabolism. Many enzymes, such as oxygenase, require oxygen for their function [3], and large amounts of energy are generated by oxidative phosphorylation compared to glycolysis [4]. A lack of oxygen leads to hypoxia—a hallmark of many cancers that is linked to tumor progression and worse clinical outcomes for patients [5]. Hypoxia in cancer can result from the fast proliferation of tumor cells; this causes some tumor cells to be located far from oxygen-supplying blood vessels (>180 μm) [6], leading to limited oxygen diffusion. In addition, the TME often promotes angiogenesis, which can result in the formation of abnormal, closed blood vessels, further inducing hypoxia [7]. Critically, the presence of hypoxia reprograms tumor cells through multiple proteins, such as hypoxia-inducible factor (HIF)-1α. Such hypoxia-adapted tumor cells are more invasive and resistant to therapies, and they can also evade immunosurveillance.Beyond tumor cells themselves, tumor-infiltrating immune cells are also under hypoxic conditions in the TME. Normoxic ambient air contains 21% O2, whereas O2 concentrations of 2–9% (14.4–64.8 mmHg) are present in tissue under physiological normoxia. In contrast, some parts of normal organs, such as the bone marrow and thymus, as well as the TME, are hypoxic, containing approximately 1% O2 (7.2 mmHg) [8]. A number of studies have investigated immune cell function in the context of hypoxia; however, these have reported contradictory results regarding whether hypoxia promotes beneficial or adverse effects on immune cells. For example, CD8+ T cells activated under hypoxia show stronger cytotoxic effects against tumor cells than those activated under normoxia [9]. Similarly, natural killer (NK) cells cultured under hypoxic conditions following normoxia are more highly activated than normoxic NK cells [10]. Conversely, hypoxia downregulates interferon (IFN)-γ production by CD8+ T cells under T cell receptor (TCR) stimulation and attenuates NK cell-mediated cytotoxicity [11,12]. Intriguingly, NK cell activity is enhanced by short-term, but not long-term, hypoxia [13,14], and continuous activation of CD8+ T cells under hypoxia promotes characteristics distinct from those present in acutely activated CD8+ T cells [15]. This suggests that while a HIF-1α-mediated glycolytic burst enhances the activity of cytotoxic cells, mitochondrial dysfunction in response to long-term hypoxia attenuates cytotoxic and inflammatory functions. Overall, these complex hypoxia-associated phenotypes are dependent on numerous different factors and experimental conditions, suggesting that understanding the role of hypoxia in antitumor immunity is likely to be more complicated than expected.Of all the organs in the body, the brain requires the highest amount of oxygen. Thus, although the brain comprises only 2% of total body weight, it consumes 20% of the body’s oxygen. Oxygen levels in the brain differ depending on the region. For example, the oxygen level in the midbrain is approximately 0.5%, whereas in the pia, it is 8% [16]. Brain tumors have been shown to contain 1.25% O2, with the peritumoral area containing 2.5% O2 [17]. Thus, brain tumor regions are mostly hypoxic compared to normal brain tissue. Glioblastoma multiforme (GBM), the most aggressive brain tumor type, can be classified as a hypoxic tumor. Irradiation is the most frequently used antitumor therapy for brain tumors; however, hypoxia-mediated stemness promotes cellular resistance to irradiation. Thus, studies aimed at understanding the effects of hypoxia within brain tumors are urgently needed. In this review, we discuss recent findings on the role of hypoxia in tumor biology and in antitumor immunity against brain tumors.Tumors of the central nervous system (CNS) are relatively rare compared to other peripheral tumors, with CNS tumors showing an estimated frequency of about 1% amongst those detected in all tumor sites [18]. Most CNS tumors, about 70% of cases, are non-malignant, half of which are meningiomas. In 30% of cases, however, CNS tumors are malignant, with glioblastomas accounting for about 50% of all CNS malignancies [19]. Despite its low overall incidence, glioblastoma is an important tumor type due to the high average years of life lost from this cancer, which amounts to 20.1 years [20]. In addition, according to a report from the Australian Institute of Health and Welfare (AIHW), between 1988–1992 and 2013–2017, the estimated percentage change in the 5-year survival rate for brain tumor was only about 2.3%, whereas all cancers combined showed an increase of 18.3% [21].Most GBMs are primary tumors, although a small portion develop from low-grade astrocytoma and thus are known as secondary GBM [22]. A study by Verhaak et al. [23] further reported that GBM can be divided into four subtypes based on gene expression patterns: classical, proneural, neural, and mesenchymal. The classical type includes amplifications of chromosome 7 and epidermal growth factor receptor (EGFR), as well as a homozygous deletion spanning the Ink4a/ARF locus. The mesenchymal type shows high expression of chitinase-3-like protein 1 (CHI3L1) and tyrosine-protein kinase Met (c-Met), with neurofibromatosis type 1 (NF1) mutation/deletion or low expression of NF1. Mesenchymal type GBM also displays a higher percentage of necrosis and inflammation than other types. The proneural type has platelet-derived growth factor receptor A (PDGFRA) abnormalities and mutations on both tumor protein p53 (TP53) and isocitrate dehydrogenase 1 (IDH1), similar to secondary GBM. In contrast, the neural type is similar to normal brain tissue in terms of the gene expression pattern. Recently, however, the Verhaak group suggested that the neural GBM subtype might, in fact, be contamination from normal brain tissue [24]. Because IDH mutation is associated with better prognosis, proneural subtype GBM is thought to be associated with better patients outcomes, whereas mesenchymal subtype GBM is associated with worse outcomes [25].To improve the survival of GBM patients, maximal resection of tumor tissues is recommended. This reduces mass effect and enhances efficacy of adjuvant therapies, leading to increased survival rates [26]. However, despite the rapid development of improved detection methods, complete resection is often difficult due to the presence of complex vasculature, location of the tumor, and fear of damage to intact brain tissues. Thus, in many cases, resection is ineffective, and recurrence after surgery is common [27].Current standard care for GBM is Stupp’s regimen, which involves radiotherapy (2 Gy per day, 5 days per week, up to a total of 60 Gy), with concomitant temozolomide (TMZ) treatment (Figure 1a) [28]. Similar to other chemotherapies, TMZ induces DNA damage via methylation of O6 and N7 positions on guanine and the N3 position on adenine, which promotes cell death. Sensitivity to TMZ treatment is largely dependent on methylation of the O6-methylguanine-DNA methyltransferase (MGMT) promoter [29]. MGMT repairs DNA, and consequently, patients with MGMT methylation, which inhibits expression of this gene, are sensitive to TMZ treatment [30]. In 2015, a device delivering alternating electric fields, also known as tumor-treating fields (TTFs), was approved by the US Food and Drug Administration for GBM patients [31]. In phase 3 trials, this device promoted a significant increase in overall survival (OS) and progression-free survival (PFS) compared to TMZ alone [32].Complex vasculature is also highly associated with GBM progression, and therefore, anti-angiogenesis therapies, including bevacizumab, have been tested. In phase 3 trials, bevacizumab treatment improved PFS; however, disappointingly, a significant improvement in OS was not observed [33,34]. Furthermore, long-term treatment with bevacizumab is associated with increased hypoxia and invasiveness [35]. Alternatively, vessel normalization via TIE2 activation and angiopoietin-2 (ANG2) inhibition showed promising results in rodent models, leading to less hypoxia and invasion, although this strategy has not been tested in humans [36].In recent years, the emergence of cancer immunotherapy was expected to be a game changer for treatment of various tumor types, including GBM. Toward this goal, many strategies have been suggested, such as dendritic cell (DC) vaccines, immune checkpoint inhibitors, chimeric antigen receptor (CAR) T cell therapy, and adoptive cell therapy. Although none of these have shown promising results for GBM patients, a massive number of studies remain ongoing [37]. Currently, the most popular strategy for GBM immunotherapy is immune checkpoint blockade, such as with anti-programmed cell death 1 (PD-1) therapy. However, recent phase 3 clinical trials using nivolumab in unmethylated-MGMT GBM with radiotherapy and methylated-MGMT GBM with chemoradiotherapy did not show any improvement in OS and PFS [38]. Although it is too early to judge, many concerns regarding the use of immunotherapy for GBM exist due to the fact that GBM is classified as a “cold tumor” with poor immune cell infiltration [39]. Thus, various strategies for converting GBM into “hot” tumor have been suggested, including the use of oncolytic viruses and/or manipulating meningeal lymphatics (Figure 1b) [40,41].GBM tumors display unique features, such as necrotic foci, pseudopalisades, and microvascular hyperplasia (Figure 2a), which are thought to be important for the fast growth and invasiveness of GBM cells [42]. Pseudopalisades may result from the migration of tumor cells escaping from a hypoxic region to form the invasive front edge. Pseudopalisading cells shape microvascular hyperplasia, forming tuft microaggregates around the edge of blood vessels and leading to the formation of glomeruloid bodies [43]. These features are largely mediated by angiogenesis-induced hypoxia. For example, excessive expression of vascular endothelial growth factor (VEGF) induces the hyper-proliferation of endothelial cells, resulting in defective and permeable blood vessels that can be easily disrupted [44]. This abnormal vasculature in the GBM microenvironment inhibits the delivery of oxygen, as well as drugs and immune cells [36]. In addition, hypoxia resulting from abnormal vessels promotes the invasion of tumor cells, a main hurdle for therapies against GBM [42].Cells can sense the surrounding oxygen level through multiple molecular mechanisms. Of these, the most well-studied is the highly conserved HIF pathway [42], which acts as the major oxygen-sensing pathway in metazoan species [45]. The transcription factor HIF is a heterodimer formed from two distinct subunits, HIFα and HIFβ. In humans, HIFα has three isoforms. HIF-1α is ubiquitously expressed and overexpressed by tumor cells [46,47]. In contrast, HIF-2α is expressed in distinct cell populations, such as in subsets of tumor-associated macrophages (TAMs) [48]. HIF-3α is also selectively expressed, although its expression in immune cells is not clear [49,50]. Target genes for HIF-1α and HIF-2α show some degree of overlap; however, a subset of genes is distinctly regulated by each transcription factor [51]. HIF-3 can also function as a transcriptional activator for a unique set of genes [52], although it is most commonly known to be a dominant-negative regulator of HIF-1, due to its lack of a C-terminal (C-TAD) domain [53]. In contrast, the aryl hydrocarbon receptor nuclear translocator (ARNT), HIF-1β is expressed ubiquitously [42].HIF-1α, HIF-2α, and HIF-1β all contain a basic helix–loop–helix (bHLH) domain, a Per–Arnt–Sim (PAS) domain, and a C-TAD domain. HIF-α also has additional oxygen-dependent degradation domain (ODDD) and an N-terminal (N-TAD) domain. The bHLH and PAS domains form the heterodimer and bind to hypoxia-response elements (HREs) [54], whereas the C-TAD and N-TAD domains are involved in transactivation of coactivators, such as p300/CBP (Figure 2b) [55].In normoxia, HIFα is bound to prolyl hydroxylase 1–3 (PHD1–3) via the ODDD, and PHD hydroxylates two prolyl residues of HIFα. PHD is regulated by O2 levels due to its 2-oxoglutarate-dependent and iron-dependent dioxygenase domains [56]. Hydroxylation of HIFα allows it to bind von Hippel–Lindau (VHL), which recruits E3 ubiquitin ligases. These promote the ubiquitination of HIFα and its subsequent degradation by the proteasome [57]. Conversely, under hypoxic conditions, PHD activity is lost, and stable HIFα translocates into the nucleus, where it binds HIFβ and coactivators, such as p300/CBP (Figure 2c) [58]. Factor-inhibiting HIF (FIH), an O2-dependent hydroxylase, also functions in HIFα regulation by blocking HIFα binding to coactivators [59]. HIF target genes are reviewed in detail elsewhere (e.g., [60]) (Figure 2c).As noted above, oxygen-dependent gene expression is mainly mediated by HIF and downstream HRE genes. The HIF-dependent hypoxic response regulates multiple cellular activities, including metabolism, migration, angiogenesis, and differentiation [61], and this pathway promotes invasiveness in hypoxic GBM cells via multiple mechanisms. For example, carbonic anhydrase 9 (CA9), a zinc-dependent enzyme that catalyzes the conversion of CO2 into bicarbonate, is known to be affected by hypoxia and is highly expressed in GBM cells [62]. Hypoxia also stabilizes the EGFRvIII protein by promoting interaction with integrin β3 in GBM cells [63] and further induces recruitment of the integrins αvβ3 and αvβ5 to the surface of GBM cells, leading to activation of focal adhesion kinase (FAK) [64]. Procollagen-lysine 2-oxoglutarate 5-dioxygenase 2 (PLOD2), an enzyme that regulates collagen cross-linking, is also controlled by hypoxia in a HIF-1α-dependent manner [65]. Collectively, these protein interactions promote invasion of GBM cells. In addition, epithelial-to-mesenchymal transition (EMT)- and metastasis-related genes, such as recombination signal binding protein for immunoglobulin kappa J (RBPJ) [66], zinc finger E-box-binding homeobox 1 (ZEB1) [67], and Twist-related protein 1 (TWIST1), are known to be regulated by hypoxia via the HIF-1α pathway [68]. Hypoxia has also been reported to induce expression of C-X-C chemokine receptor type 4 (CXCR4) in GBM cells and of CXCL12 in endothelial cells, and both C-C motif chemokine receptor type 5 (CCR5) and C-C motif chemokine ligand 4 (CCL4) are positively regulated by hypoxia [69,70,71].Several key pathways are involved in cellular adaptation to hypoxia. For example, hypoxia suppresses cap-dependent protein translation at the level of translation initiation [72]. This suppression is primarily regulated by protein kinase R (PKR)-like endoplasmic reticulum (ER) kinase (PERK) and the mechanistic target of rapamycin (mTOR) complex 1 (mTORC1) [73]. However, some genes are continuously translated under hypoxia, such as those associated with stress response. These commonly include genes related to antioxidant response, amino acid transport, metabolism, and autophagy. Further, when activated PERK phosphorylates eIF2α to induce translational suppression [74], the remaining ribosomes are able to translate mRNAs encoding proteins for unfolded protein response (UPR), such as ATF4 [75]. In addition, activation of inositol-requiring transmembrane kinase/endoribonuclease 1α (IRE1α) in response to hypoxia promotes activation of functional X-box binding protein 1 (XBP1), which regulates multiple metabolic pathways [76,77]. Mitochondrial functions are also regulated by hypoxia, as 0.3% O2 is the rate-limiting threshold for electron transport complex (ETC) activity [78]. Likewise, the tricarboxylic acid (TCA) cycle is reduced, mitochondria translocate to the perinuclear site, and mitochondrial fission and mitophagy are induced in both a HIF-dependent and HIF-independent manner [79,80,81,82]. The most prominent aspect of adaptation to hypoxia is upregulation of glucose uptake and glycolysis. This is mediated by HIF-1α, which directly regulates expression of glucose transporter 1 (GLUT1), GLUT3, hexokinase 1 and 2, enolase 1, phosphoglycerate kinase 1 (PGK1), pyruvate kinase M2 (PKM2), lactate dehydrogenase A (LDHA), and phosphoinositide-dependent kinase 1 (PDK1) [83,84]. PDK1 inhibits conversion of pyruvate to acetyl CoA, thereby promoting lactate production [85]. This glycolysis-mediated enrichment of lactate and H+ ions lowers the surrounding pH, and critically, both the presence of lactate and low pH are harmful for antitumor immunity [86]. Hypoxia adaptation mechanisms are also involved in cell death pathways. For example, hypoxia induces autophagic cell death in apoptosis-competent cells via BCL2-interacting protein 3 (BNIP3) [87] and promotes necrosis of neuronal cells [88]. However, alarmin release by necrotic cells further promotes progression of glioblastoma stem-like cells [89].Critically, hypoxia is also associated with increased radioresistance in GBM. Although the underlying mechanism is not clear, several molecular pathways have been implicated in this phenomenon. In one instance, it was shown that the mitogen-activated protein kinase (MAPKK; MEK)/Extracellular signal-regulated kinase (EKR) pathway promotes hypoxia-mediated radioresistance via the activity of DNA-dependent protein kinase, catalytic subunit (DNA-PKcs) and HIF-1α [90]. Another study found that phospholipase C gamma (PLCγ) binding to fibroblast growth factor receptor 1 (FGFR1) induces protein kinase C (PKC) activation in response to HIF-1α regulation, and this also induces radioresistance [91]. In addition, hypoxia promotes glioma stem cell (GSC) formation by inducing stem cell marker genes, including octamer-binding transcription factor 4 (OCT4), NANOG, SRY-Box transcription factor 2 (SOX2), Kruppel-like factor 4 (KLF4), and cMYC, while downregulating expression of glial fibrillary acidic protein (GFAP) [92,93], and this was further shown to be critical for inducing radioresistance.HIF proteins and HRE genes also regulate a number of angiogenesis-related molecules, such as VEGFs, placenta growth factors (PGFs), angiopoietin (ANGPT), CXCL12, and platelet-derived growth factor B (PDGF-B) [94]. In response to hypoxia, VEGFs and PGFs bind to VEGF receptor (VEGFR)-1 and -2 on endothelial cells and induce proliferation and survival via the ERK/PI3K/AKT pathways [95]. Rho GTPase-mediated migration and membrane type matrix metalloproteinase (MMP)-mediated extracellular matrix (ECM) degradation are also induced by VEGF/PGF binding [96,97]. Among the angiogenesis-related molecules induced by hypoxia, the most well-studied protein is VEGF-A. Notably, although this protein is critical for homeostatic vasculature, hypoxia-mediated VEGF-A overexpression induces vascular permeability, which in turn, inhibits the delivery of drugs and immune cells, limits perfusion, and even further promotes hypoxia [94].Like other cells, immune cells need proper oxygen levels for survival and function. However, despite mechanisms to maintain homeostasis and normoxia in most instances, some niches can become hypoxic due to anatomic characteristics of organs or burst of cellular expansion. This is referred to as “physiological hypoxia” [50], and in some cases, it is necessary for proper organ function. The bone marrow is one of the most well-characterized physiologically hypoxic organs [98]. Here, hypoxia is critical for maintaining hematopoietic stem cell (HSC) homeostasis [99]. Although it remains controversial, HIF-1β was shown to be required for quiescence, survival, and development of HSCs, whereas HIF-2α is dispensable for HSC function [100,101]. Germinal centers (GCs) are another example of a physiologically hypoxic environment. During maturation of B cells, the oxygen gradient decreases within the GC [102], possibly as a result of increased oxygen consumption by expanding B cells. Notably, GC hypoxia was found to affect phenotype, proliferation, and class switching of B cells [102,103]. The reproductive organs also show physiological hypoxia. For example, although exact O2 tension values in seminiferous tubules remain controversial, the testicular interstitium is hypoxic, showing O2 tension values of about 12 to 15 mmHg [104]. Likewise, the vagina is hypoxic under normal conditions [105], and physiological hypoxia in the placenta modulates immune function by protecting the fetus from the maternal immune system; accordingly, HIF dysfunction is associated with placental defects [106]. HIF-1α-mediated gene expression in trophoblasts regulates non-classical class I histocompatibility antigen to prevent attack from natural killer (NK) cells [107], and HIF-1α-mediated programmed cell death 1 ligand 1 (PD-L1) was further found to inhibit T cell responses [108]. The intestinal mucosa is also hypoxic [109], and here, it was shown that physiological hypoxia regulates epithelial barrier function and resident immune cells [110].In general, antitumor immunity is similar to persistent antiviral immunity [111]. Although many different immune cells can participate in antitumor responses, T cells are thought to be the most important antitumor immune cells. When tumor antigen is released, antigen-presenting cells (APCs), such as DCs, take-up antigen and migrate into the lymph nodes (LN), where they present their antigen to T cells. Antigen-specific T cells then undergo priming and clonal expansion, and these activated T cells migrate into the tumor area. CD4+ T cells orchestrate antitumor immunity and CD8+ T cells recognize and directly kill tumor cells [112]. However, tumor cells often poorly express major histocompatibility complex (MHC) class I molecule to escape from CD8+ T cell-mediated immunosurveillance [113]. In addition, most tumors have antigens that resemble our self-antigens, thereby inducing tolerance [114]. This limits antitumor T cell activity, and as a result of persistent activation, T cells become exhausted and lose their function. Exhausted CD8+ T cells, for example, will express PD-1 on the surface, which binds to PD-L1 on tumor cells or myeloid cells and inhibits T cell function [115,116].Another class of T cells known as γδ T cells also participate in antitumor immunity and are highly correlated with favorable patient outcomes [117]. These cells can recognize tumor cells via γδTCR or NK-like receptors [118], the upregulation of their ligands of which are induced by transformation and cellular stress [119]. For example, tumor cells highly express NK group 2 member D (NKG2D)-ligands, such as MHC class I polypeptide-related sequence A/B (MICA/B) in humans and retinoic acid early transcript 1 (RAE-1) for mice [120].Beyond T cells, myeloid cells, including macrophages, monocytes, and neutrophils, can also act as antitumor cells via phagocytosis or the production of inflammatory cytokines [121,122]. However, most tumor-infiltrating myeloid cells are immunosuppressive; these are known as myeloid-derived suppressive cells (MDSCs). MDSCs can suppress antitumor immunity through multiple mechanisms, including interleukin (IL)-10 secretion [123]. Likewise, regulatory T cells (Tregs) also suppress antitumor immune responses [124], and various other immune cells, such as innate lymphoid cells (ILCs) [125], B cells [126], and eosinophils [127], can promote antitumor or protumor responses, depending on the context and environment.Unlike peripheral tumors, antitumor immunity against brain tumors has been poorly described. Because of the strong blood–brain barrier (BBB), infiltration of lymphocytes is limited, and thus, the brain is considered to be an immune-privileged organ [128]. Microglia are the predominant immune cells in the brain, although a limited number of other immune cells, such as T cells and mast cells, are also present [129,130]. Likewise, the brain tumor microenvironment is also primarily enriched with microglia and bone marrow-derived macrophages [131]. Due to these unique characteristics, as well as low frequency of neoantigen, antitumor immunity in the brain and responsiveness to immunotherapies is quite poor (Figure 3) [132]. Thus, brain tumors such as GBM are often referred to as “cold tumors” [133].However, a recent study found that that classical DC-1s (cDC1s) can infiltrate into the GBM area and present antigen to T cells in deep cervical LNs (dcLNs) [134]. This study further suggested that CD141+ cDC1s can present antigen in human GBM patients. Regardless, the role of cDC1s and CD8+ T cells in GBM is negligible without immunotherapies, such as anti-PD-L1 treatment [135]. Further, our group showed that CD4+ and CD8+ T cells are dispensable for OS of GBM patients and animals [136]. These studies suggest that although immune responses do occur in the GBM microenvironment, they are too weak to protect host. Consistent with these observations, Song et al. showed that meningeal lymphatics are dampened by GBM progression, suggesting this is one reason for poor anti-GBM immunity [41]. Notably, they further showed that if lymphatics are improved by VEGF-C application, most GBM-bearing animals can survive (Figure 3) [41]. However, extracranial antigen presentation was found to be unable to promote tumor eradication without immunotherapies in a melanoma brain metastasis model [137]. This suggests that antigen presentation in the periphery is unlikely to be sufficient for inducing anti-brain tumor immunity, and further study is needed.Macrophages and microglia are also important components of anti-GBM immunity. Past in vitro studies have suggested that macrophages can be divided by two groups: M1 and M2. M1 macrophages are related to Th1 responses, whereas M2 macrophages regulate wound healing and Th2 responses [138]. However, recent studies have shown that subsets of macrophages are more complex than previously thought. Tumor-associated macrophages (TAMs), for example, are neither M1- nor M2-like, and rather show mixed phenotypes [139]. Within the tumor microenvironment, most TAMs are M2-like cells; however, proinflammatory TAMs that can phagocytose tumor cells also exist [140]. One well-known mechanism by which this occurs is via the SIRPα–CD47 axis. CD47 is a “do not eat me” signal that inhibits phagocytosis of SIRPα-expressing TAMs. Thus, we can improve phagocytosis by TAMs using CD47 blockade [141]. In addition, the depletion of M2-like TAMs by blocking colony stimulating factor 1 receptor (CSF1R) is also considered to be a promising therapeutic strategy [142]. Conversely, although they are dominant type of immune cells in the brain, the role of microglia in GBM is still unclear. Microglia are usually located at the surrounding edge of the tumor mass rather than inner area [143], but the reason for this is not known, and more studies on microglia in GBM are needed.NK cells and γδ T cells are also able to kill GBM cells [144,145]. Notably, although in vivo blocking of NK1.1 does not affect OS of GBM-bearing mice [136], another study showed that NK1.1-blockade increases GBM size [146], suggesting an antitumor role for NK cells. Intriguingly, anti-GBM NK cell activity was found to be dependent on the gut microbiota [146]. In addition, NK cells display more potent activity against stem-like GBM cells [144], although direct contact between NK cells and GBM stem cells (GSCs) via αv integrin induces TGF-β-mediated NK cell suppression [147]. Thus, more studies on the role of NK cells in GBM are needed. Other groups have focused on the role of γδ T cells; one murine study showed that Vγ1, Vγ4, and Vγ7 T cells are present in the brain tumor area [148], with Vγ7 comprising the most dominant γ chain. This study further showed that Vδ1, Vδ4, and Vδ6.3 T cells are able to infiltrate into GBM tissue, and in this case, Vδ6.3 is the most prevalent δ chain. In addition, findings from this investigation suggested that γδ T cells are diminished at the terminal stage of tumor progression due to apoptosis, with Vδ6.3+ T cells showing the most vulnerability to cell death. Preferential infiltration of Vγ9Vδ2 T cells was also observed in human brain tumor tissue [149], and interestingly, these cells can preferentially kill mesenchymal GBM cells via NKG2D [150]. Further, despite limited investigation, one study suggested that B cells are immunosuppressive within the GBM [151].Hypoxia has been shown to affect multiple functions of anti-cancer immune cells. For example, in vitro culture of CD8+ T cells under hypoxic conditions promotes reduced levels of proliferation, cytokine production, and cytotoxicity. Melanoma-infiltrating CD8+ T cells are also highly exhausted and malfunctional due to severe hypoxia [11]. Similarly, immune cells in GBM core tissue, which primarily comprise M2 macrophages and regulatory T cells, are highly hypoxic (Figure 4a). CD8+ T cells within the hypoxic core are also exhausted, and peripheral CD8+ T cells cultured in vitro under hypoxic conditions phenocopy CD8+ T cells from the GBM core [152].Expression of HRE genes, including PD-L1, is induced by hypoxia in tumor cells and immune cells [153]. However, in melanoma, hypoxia-induced metabolic stress also inhibits mitochondrial biogenesis in tumor-infiltrating lymphocytes (TILs). Competition between tumor cells and TILs further suppresses the metabolic activity of TILs and inhibits reactivity against immune checkpoint blockade [154], and these immunosuppressive effects are enhanced if hypoxia is chronically persistent [15]. Notably, although PD-1 blockade alone is unable to rescue mitochondrial dysfunction, metabolic reprogramming is sufficient to reverse the exhaustion of TILs [155]. However, TILs in the GBM may be different from those present in subcutaneous tumor models. One study found that modulation of hypoxia using metformin is not sufficient to reinvigorate CD8+ T cell responses in a GBM model [136], and further study is needed.CD4+ T cells include various subsets; T helper 1 (Th1) cells resemble CD8+ T cells, whereas Tregs show weaker glycolysis and are more oxidative than effector T cells [156]. One study using a B16 melanoma model reported that glucose uptake is closely related to Treg stability, as these cells utilize lactic acid to stabilize their suppressive identity [157]. In the GBM microenvironment, oxidative phosphorylation promotes immunosuppression of Tregs, whereas glycolysis enhances migration (Figure 4a). Further, in a hypoxic microenvironment, Tregs in the GBM use fatty acids for immunosuppression [158], and this is tightly regulated by HIF-1α.Because hypoxia transiently disrupts the BBB, it may be related to inflammation and inflammation-associated features of the microglia [159]. In Alzheimer’s disease (AD), acute hypoxia induces the M1 transition of microglia [160]. However, hypoxia also inhibits mitochondrial metabolism and promotes the cell cycle arrest of microglia in AD [161]. In GBM tissue, microglia are highly distributed near pseudopalisades and do not escape from hypoxia. In addition, microglia under hypoxia show elongated morphology and increased phagocytosis capability (Figure 4b) [162]. However, the precise role of microglia and hypoxic microglia in GBM is unknown and requires further investigation. Intriguingly, recent studies have shown that other brain-resident cells can participate in GBM progression. For example, oligodendrocyte precursor cells (OPCs) may be associated with GBM progression, and these cells have been proposed as a possible origin of GBM cells [163]. Synaptic and electric communication between GBM cells and neurons also promotes GBM progression [164]. In addition, it was shown that astrocytes suppress immune response in the GBM microenvironment [165]. Because these cell types are also affected by hypoxia, this should be further studied in the context of GBM. Intriguingly, one study reported that neuronal expression of PD-L1, which is known to be regulated by hypoxia, is related to better prognosis of GBM patients [166], thus suggesting a possible role for hypoxic brain cells in GBM progression (Figure 4b).Hypoxia is also related to dysfunction of NK cells (Figure 4c) [167], as hypoxia-associated mitochondrial fragmentation disrupts NK cell-mediated antitumor immunity [168]. However, the precise effect of hypoxia on NK cell function within the GBM microenvironment remains unclear. Similarly, cytotoxicity of γδ T cells, which are similar to, but distinct from, NK cells, and can kill tumor cells via both γδTCR and NKG2D, is also dampened by hypoxia (Figure 4c) [169]. Further, in the GBM microenvironment, γδ T cells are apoptotic and malfunctional [148], due to the fact that GBM patients receive radiochemotherapy, which also kills γδ T cells. Thus, MGMT-modified γδ T cell therapy might represent an alternative treatment strategy [170]. However, in a murine model, γδ T cells were found to be apoptotic without any treatment [148]. Thus, we expect that the tumor microenvironment suppresses γδ T cell function. Results from our previous study suggest that hypoxia is the critical mechanism mediating suppression of γδ T cells in the GBM [136]. Specifically, we found that when metformin is given to GBM-bearing mice, tumor cell respiration is inhibited, and the remaining oxygen can be utilized by γδ T cells. In addition, adoptive γδ T cell therapy with metformin or HIF-1α is able to prolong overall survival of GBM-bearing mice. Thus, rescuing hypoxia of γδ T cells could be beneficial for γδ T cell-mediated anti-GBM immunity.GBM is one of the most immunologically poor tumor types. Although a number of mechanisms, such as the BBB, participate in immunosuppression, hypoxia is a critical immunosuppressive factor in the GBM microenvironment. Thus, even when immune cells are able to infiltrate into the GBM microenvironment, this hypoxic niche suppresses their antitumor functions. As discussed above, hypoxia inhibits multiple immune cells that are important for antitumor immunity, including CD8+ T cells and γδ T cells. Conversely, functions of immunosuppressive cells, such as Tregs and M2 macrophages, are enhanced by hypoxia. As a consequence of this strong immunosuppression, clinical trials assessing the use of immunotherapy for the treatment of GBM have been unsuccessful. An improved understanding of the unique features of anti-GBM immunity is therefore urgently needed to overcome these hurdles and develop effective treatment options. In particular, studies aimed at further investigating the effects of hypoxia on the multiple types of GBM-infiltrating immune cells will help to elucidate the mechanisms by which hypoxia suppresses immune function and determine how this could be overcome. One possible approach is through the use of immunotherapy combined with anti-hypoxic strategies, such as vessel normalization. Alternatively, cell therapy using engineered hypoxia-resistant immune cells may be another option for next-generation immunotherapy against hypoxic tumors.Hypoxia is a classic hallmark of tumors [5], and GBM is one of the most hypoxic tumors. Hypoxia affects multiple aspects of GBM biology and pathology, including vasculature, invasiveness, resistance to drugs, and antitumor immune responses [42,171]. Critically, hypoxia-driven invasion and angiogenesis are highly associated with poor prognosis, and hypoxia-mediated resistance to conventional therapies, including chemotherapy and radiation, is a significant hurdle when caring for GBM patients. Furthermore, recently developed immunotherapies are also not effective against GBM, as it is considered to be a “cold tumor”, with low neoantigen levels and poor immune cell infiltration [132]. There are many reasons why GBM is a “cold tumor”, and hypoxia is one critical factor that promotes immunosuppression within the GBM microenvironment [61]. Although the detrimental role of hypoxia in immune cell function has been well-studied, the precise impact of hypoxia on anti-GBM immunity is unclear. In particular, CD8+ T cells are known to be suppressed by hypoxia, but unlike in other tumors, a re-oxygenation strategy was not effective for restoring the CD8+ T cell function in GBM. In contrast, improving the oxygen metabolism of γδ T cells was sufficient to increase the survival period of GBM-bearing animals, suggesting a critical role for these cells [136]. Thus, the effect of hypoxia on various immune cell types within the GBM is a critical area of investigation, as targeting hypoxia may be beneficial for improving the efficacy of conventional therapy and immune responses against this deadly tumor.Conceptualization, J.H.P. and H.K.L.; writing—original draft preparation, J.H.P. and H.K.L.; writing—review and editing, J.H.P. and H.K.L.; visualization, J.H.P. and H.K.L.; supervision, H.K.L.; project administration, J.H.P. and H.K.L.; funding acquisition, J.H.P. and H.K.L. All authors have read and agreed to the published version of the manuscript.This study was supported by the National Research Foundation of Korea (NRF-2021M3A9D3026428 and NRF-2021M3A9H3015688) funded by the Ministry of Science and ICT of Korea.The authors thank the members of the Laboratory of Host Defenses for helpful discussions. Figures were created with BioRender.com.The authors declare no conflict of interest.Therapeutic approaches for glioblastoma multiforme (GBM) and hurdles to treatment. (a) There are multiple strategies to care for GBM patients. Conventional GBM therapy involves surgery, followed by radiotherapy and concomitant chemotherapy. However, this is not fully effective. Recently, immunotherapies have been developed and shown promising results for other tumors. Immune checkpoint blockade approaches for inhibiting immunosuppression, and cell therapies, such as dendritic cell (DC) vaccines, are now being tested for GBM. However, responsiveness to these therapies is poor. (b) Tumors can be classified as “hot” or “cold”. Hot tumor shows high levels of neoantigens, increased infiltration of immune cells, and better responsiveness to therapies relative to cold tumors. Thus, several approaches for converting cold tumors into hot tumors, such as by manipulating lymphatics or through the use of oncolytic viruses, are being studied.Responses to hypoxia in GBM tumors. (a) GBM tissue shows aggressive invasiveness and pseudopalisades. Oxygen is supplied by blood vessels, and thus, tumor cells located far from vessels become hypoxic due to poor oxygen diffusion, forming necrotic core (foci). Tumor cells that escape from hypoxia form pseudopalisades. (b) Hypoxia-inducible factor (HIF)-α and HIF-β have basic helix–loop–helix (bHLH), Per–Arnt–Sim (PAS), and C-terminal (C-TAD) domains. The bHLH and PAS domain are responsible for forming the heterodimer, and the C-TAD domain promotes transactivation of co-activators. HIFα also has an N-terminal (N-TAD) domain and an oxygen-dependent degradation domain (ODDD), which mediate its oxygen-dependent degradation via the ubiquitin–proteasome pathway. The N-TAD also participates in transactivation of co-activators. (c) Under normoxia, oxygen-dependent prolyl hydroxylase (PHD) enzyme is active and binds to HIF in an ODDD-dependent manner. PHD hydroxylates HIF, which allows von Hippel–Lindau (VHL) to bind and recruit E3 ubiquitinase. This enzyme ubiquitinates HIF, targeting it for binding and degradation by the proteasome. In contrast, under hypoxia, HIF is stable and translocates into the nucleus, where it binds to co-activators, such as p300/CBP or TIP60, and turns on expression of hypoxia-response element (HRE) genes. HIF can regulate multiple cellular processes via activation of these HRE genes. HIF-suppressors, including factor-inhibiting HIF (FIH), function to inhibit binding between HIF and its co-activators and block HRE activation.Immune responses in the GBM. GBM antigens are drained by meningeal lymphatics, and classical DC-1s (cDC1s) present antigen to CD8+ T cells in the deep cervical lymph node. However, due to the strong blood–brain barrier (BBB), immune cell infiltration into the parenchyma is limited. In addition, the predominant microglia suppress immune responses. However, enhancing lymphatics via VEGF-C treatment has shown promising results to improve survival in animal models.The effect of hypoxia on anti-GBM immune responses. (a) Under hypoxia, macrophages preferentially show an immunosuppressive M2-like phenotype, rather than an inflammatory M1-like phenotype. Tumor-infiltrating lymphocytes are suppressed in response to activation of HRE genes and mitochondrial dysfunction, and accumulation of lactic acid supports stability and function of regulatory T cells (Tregs), to further suppress inhibit responses. (b) Microglia in the tumor area show an elongated phenotype; these cells are immunosuppressive, with enhanced phagocytosis ability. Oligodendrocyte precursor cells are thought to be a precursor for GBM cells. Connection with neurons also supports GBM progression, whereas PD-L1 expression from neurons is associated with improved prognosis of GBM patients. (c) Hypoxia inhibits functions of natural killer (NK) cells and γδ T cells and promotes dysfunction of NK cell mitochondria.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01177.txt

@@ -0,0 +1,3 @@
+These authors contributed equally to this work.Immune checkpoints blockade has emerged as an effective approach to prevent immune escape of tumor cells, and constitutes a powerful anti-cancer therapeutic strategy. Regulation of the expression of genes encoding immune checkpoint inhibitors has thus become an increasingly important field of study. Beyond transcription, gene expression is regulated at several post-transcriptional levels including pre-mRNA 3′-end processing and mRNA translation. More specifically, the eIF4F translation initiation complex represents an important hub for oncogenic signaling in the etiology of different cancers. The eIF4A RNA helicase component of the eIF4F can be inhibited by the widely characterized small molecule inhibitor silvestrol. Here, we evaluated the effect of eIF4A inhibition with silvestrol on the translation of alternatively polyadenylated mRNAs in melanoma cell lines and activated T cells. We show that silvestrol can selectively inhibit the translation of alternatively polyadenylated isoforms of genes encoding key immune-related proteins.Targeting the translation initiation complex eIF4F, which binds the 5′ cap of mRNAs, is a promising anti-cancer approach. Silvestrol, a small molecule inhibitor of eIF4A, the RNA helicase component of eIF4F, inhibits the translation of the mRNA encoding the signal transducer and activator of transcription 1 (STAT1) transcription factor, which, in turn, reduces the transcription of the gene encoding one of the major immune checkpoint proteins, i.e., programmed death ligand-1 (PD-L1) in melanoma cells. A large proportion of human genes produce multiple mRNAs differing in their 3′-ends through the use of alternative polyadenylation (APA) sites, which, when located in alternative last exons, can generate protein isoforms, as in the STAT1 gene. Here, we provide evidence that the STAT1α, but not STAT1β protein isoform generated by APA, is required for silvestrol-dependent inhibition of PD-L1 expression in interferon-γ-treated melanoma cells. Using polysome profiling in activated T cells we find that, beyond STAT1, eIF4A inhibition downregulates the translation of some important immune-related mRNAs, such as the ones encoding TIM-3, LAG-3, IDO1, CD27 or CD137, but with little effect on the ones for BTLA and ADAR-1 and no effect on the ones encoding CTLA-4, PD-1 and CD40-L. We next apply RT-qPCR and 3′-seq (RNA-seq focused on mRNA 3′ ends) on polysomal RNAs to analyze in a high throughput manner the effect of eIF4A inhibition on the translation of APA isoforms. We identify about 150 genes, including TIM-3, LAG-3, AHNAK and SEMA4D, for which silvestrol differentially inhibits the translation of APA isoforms in T cells. It is therefore crucial to consider 3′-end mRNA heterogeneity in the understanding of the anti-tumor activities of eIF4A inhibitors.Immune checkpoint blockade is one of the most effective approaches to activate therapeutic antitumor immunity as tumors often use immune-checkpoint pathways as a major underlying mechanism of immune resistance. This involves immune receptors that negatively regulate antitumor adaptive T cell (T lymphocyte) responses, such as Cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) and programmed cell death protein 1 (PD-1) or its ligand PD-L1. Antibodies targeting these receptors are now widely used to treat a broad range of cancers. Although these new immunotherapies represent a huge improvement in the field of cancer therapies, early or late resistance emerge in the majority of the patients [1,2]. There is thus a high medical need to better understand the mechanisms underlying the control of immune checkpoint gene expression that is so far essentially described at the transcription level [3,4,5,6].Following their transcription, most eukaryotic precursor messenger RNAs (pre-mRNAs) undergo a number of nuclear processing events including (i) a 5′ end capping reaction, (ii) splicing that is the removal of introns and subsequent ligation of exons, and (iii) a 3′-end RNA cleavage followed by addition of a polyadenylated tail at a polyadenylation site (pA site) on the pre-mRNA. The 3′-polyadenylated tail of mRNAs is necessary for their transport to the cytoplasm, their stability and translation [7,8,9,10]. Alternative polyadenylation (APA), which occurs in about two-thirds of human genes, is the alternative usage of distinct pA sites in genes [11]. APA can lead to the production of mRNAs with different lengths of their 3′ untranslated region (3′UTR) or with different protein coding capacities. In the latter case (called intronic polyadenylation; IPA), an alternative pA site located upstream of the last exon of the gene is used, leading to an alternative last exon (which may or may not be annotated), and the resulting alternatively polyadenylated mRNA isoform differs not only in its 3′UTR nature but also in its carboxy-terminal coding region [12,13,14].Translational control has emerged as an important regulatory mechanism associated with many hallmarks of cancer. The eIF4F complex, composed of the 5′-cap-binding protein eIF4E, the RNA helicase eIF4A and the scaffolding protein eIF4G, is one of the most extensively studied RNA binding complexes involved in translational control in cancer [15,16,17,18]. eIF4F promotes the recruitment of the 40S ribosomal subunit to the cap. This recruitment is dependent on several features of the mRNA, including the level of RNA secondary structure in the 5′UTR, which is controlled by the unwinding activity of the eIF4A RNA helicase. Thus, specific mRNAs are more dependent on the eIF4F complex and eIF4A activity [19]. Among numerous small-molecule inhibitors of this complex reported to exert antitumor effects, silvestrol, a natural small-molecule selective inhibitor of eIF4A, is one of the most extensively studied [20,21,22]. eIF4A targeting by silvestrol selectively inhibits the translation of important oncogenic mRNAs containing a high level of secondary structure in their 5′UTR, hence exerting antitumoral effects [15,20,23,24,25,26].We recently showed that, in interferon-γ-treated melanoma cells, translation of the signal transducer and activator of transcription 1 (STAT1) transcription factor is upregulated in an eIF4F-dependent manner, leading to transcriptional upregulation of PD-L1 [16]. In fact, the STAT1 gene has two alternative pA sites that generate a short transcript encoding the STAT1β protein and a long transcript encoding the most studied STAT1α protein. More generally, many genes were recently described to have alternatively polyadenylated mRNA variants in human immune cells [27]. With the exception of a few genes [28,29,30], the differential functions of protein isoforms encoded by alternatively polyadenylated mRNAs are poorly documented. Here, we evaluated the effect of inhibiting eIF4A with silvestrol on the translation of alternatively polyadenylated mRNA isoforms in both melanoma and T cells.A375, SK-MEl-2, WM793, MCF-7, and MDA-MB-231 cells were grown in DMEM (Eurobio) containing 10% FBS (Pan Biotech, AidenBach, Germany) and L-Glutamine (Eurobio Scientific, Les Ulis, France) at 37 °C and 5% CO2. siRNA reverse transfections were carried out in 10 cm tissue culture dishes with Lipofectamine RNAiMAX (Thermo Scientific, Les Ulis, France). The siRNAs (Dharmacon, Cambridge, UK) were used at a final concentration of 30 nM; STAT1 α: TGTTATAGGTTGTTGGATA and STAT1 β: CAGAAGAGTGACATGTTTA) as per the manufacturer’s instructions in OptiMEM reduced serum media (Thermo Scientific). Isolated T cells from PBMCs (whole blood of healthy donors acquired from Établissement Français du Sang, Île de France) and Jurkat cells were cultured in RPMI (Eurobio Scientific, Les Ulis, France) supplemented with 10% (v/v) fetal bovine serum (PAN Biotech, Aidenbach, Germany), 2 mM L-glutamine (Eurobio Scientific, Les Ulis, France) and maintained at 37 °C in 5% CO2. They were stimulated with ImmunoCultTM Human CD3/CD28/CD2 T Cell Activator (StemCell #10990) and IL-2 (Peprotech 200-02) for 72 h. For protein analyses: 24 h after transfection, cells were treated with 10 nM or 30 nM silvestrol (MedChemExpress HY-13251, Sollentuna, Sweden). In addition, 48 h after transfection, cells were washed with PBS harvested on ice. For polysome experiments: Treatment with silvestrol was performed at a final concentration of 10 nM for 2 h.Cells were harvested by scraping on ice, treated with Fc Block (BD Biosciences, Rungis, France) in PBS/EDTA/BSA, washed and then incubated with the primary antibody human PD-L1 (Biolegend 329708, Paris, France) for 30 min. They were then incubated with PBS/EDTA/BSA containing Zombie NIR (live/dead discriminant) and then resuspended in 500 μL PBS/EDTA/BSA. An LSRII flow cytometer (BD Biosciences) was used for the acquisition of stained cells and the analysis of acquired data was performed using the FlowJo software. During acquisition of stained cells, debris (low FSC and SSC) was excluded and a selection of single cells that were negative for the live/dead discriminant was done by gating. An isotype negative control was used to eliminate a non-specific background signal according to the datasheet supplied with each antibody.Cells were harvested on ice in PBS and centrifuged at 400× g for 5 min at 4 °C. The cell pellets were resuspended in RIPA Buffer containing phosphatase inhibitors and protease inhibitors (EDTA-free) (Thermo Scientific, Les Ulis, France). A dosage of the protein content in the cell lysates was performed using a bicinchoninic acid protein assay kit (Thermo Scientific, Les Ulis, France). Protein samples were then loaded onto denaturing NuPAGE gels (Life Technologies, Alfortville, France), resolved and transferred to a 0.45-mm nitrocellulose membrane (Bio-Rad, Marnes-la-Coquette, France). Blocking of the membranes was carried out in a buffer containing TBS, Tween-20 and 5% milk, following which they were incubated with the appropriate antibodies. Visualization of proteins was carried out using an ECL system (Bio-Rad). Relative density was calculated for quantitative analysis of band intensities using the ImageJ software (Java 1.8.0_172, imagej.nih.gov accessed on 19 January 2022, Bethesda, MD, USA).The following primary antibodies were used: STAT1 (Cell Signaling Technology 14994) and SEMA4D (Thermofisher Scientific H00010507-M01).A fractionation of subpolysomal and polysomal ribosome fractions was performed by sucrose density gradient centrifugation. The cells in culture were collected following a treatment with 100 μg/mL cycloheximide at 37 °C. They were harvested by scraping on ice in cold PBS containing 100 μg/mL cycloheximide, centrifuged at 400× g for 5 min and then resuspended in 400 μL of hypotonic buffer (5 mM Tris, pH 7.5, 1.5 mM KCl, 2.5 mM MgCl2, complete protease and phosphatase inhibitors) containing 0.5% Triton X-100, 0.5% sodium deoxycholate, 2 mM DTT, 400 U/mL RNaseOUT and 100 μg/mL cycloheximide. The lysates were loaded onto a 5–50% sucrose density gradient and centrifuged in an SW41 rotor at 38,000 rpm for 2 h at 4 °C. The polysomal profiles were monitored, and the fractions collected using a gradient fractionation system.For the polysome profiling experiments, RNA extraction was performed using the TRIzol method (TRIzol-LS) from 250 μL of each fraction. The SuperScript IV Reverse Transcriptase (Thermo Scientific) kit was used for cDNA synthesis using random hexamer primers according to the manufacturer’s instructions. Equal volume of cDNA from each fraction was used to carry out the qRT-PCR experiments. For total RNA preparation, mRNA isolation was performed using TRIzol (Invitrogen, Paris, France) according to standard procedures. qRT-PCR was performed using the PowerUP qPCR Master Mix (Thermo Scientific) and SuperScript IV Reverse Transcriptase (Thermo Scientific) and was monitored on a Viia 7 System (Applied Biosystems, Paris, France). The HPRT gene was used to normalize the results in 2−ΔΔCt analyses. The primer sequences of each cDNA were designed using Primer-BLAST (ncbi.nlm.nih.gov accessed on 19 January 2022) (Table S1).3′-seq libraries were prepared with QuantSeq 3′ mRNA-Seq Library Prep Kit REV for Illumina (Lexogen, Vienna, Austria) using 500 ng of polysomal and input (cytosolic) RNAs (n = 3 for each condition) following manufacturer’s instructions. Purified libraries were quantified with Quant-iT Picogreen dsDNA kit (ThermoFisher Scientific) and run on Experion automated electrophoresis system (Bio-Rad). Pooled libraries were quantitated by qPCR (KAPA Library Quantification Kits Illumina Platforms, Roche), diluted to 12 pM, and subjected to single-end, 51 bp sequencing using the NovaSeq 2500 machine (Illumina).For all samples, raw reads were trimmed to remove uninformative nucleotides arising from primer sequences. Trimmed reads of 25 bp or more were aligned on the human reference genome (hg19) using Bowtie2 (version 2.2.5). [31] Mapping quality score (MAPQ) was determined and only reads with a score of 20 or more were selected (Samtools version 1.1) for downstream analysis. Clustering of reads was carried out along the genome using Bedtools (version 2.17.0) [32], allowing a maximum distance of 50 bp and a minimum number of 5 reads per peak. Peaks with a stretch of 6 consecutive adenosines (or 8 adenosines out of 9 nucleotides) within 50 bp downstream were filtered out, as they are likely due to internal priming of oligo-dT. Overlapping peaks from all samples of the conditions under comparison were merged to define a common set of genomic sequences corresponding to polyA sites. Location of peaks within genes was annotated using gene coordinates on the basis of overlapping Refseq transcripts with the same gene symbol. Peaks localized in the intronic region of a gene were classified as intronic polyA (IPA) peaks and those in the last exon of a gene were classified as LE peaks. Differential analyses between control and treated conditions were done with three independent biological replicates per condition. Comparing of the regulation of each IPA to the regulation of the gene’s last exon (taken as the sum of the peaks in this exon), we used the version 1.4.5 [33] and the following statistical model:
+		Yij = μ + Li + Cj + (LC)ij + Eij
+        where Yij is the normalized counts of peak i in biological condition j, μ is the mean, Li is the peak localization (IPA or LE), Cj is the biological condition, (LC)ij is the interaction between peak localization and biological condition, and Eij is the residual. p-values and adjusted p-values (Benjamini–Hochberg) were calculated. Data with p < 0.05 are shown. The complete bioinformatics pipeline (3′-SMART package) explained in this paper is freely available at GitHub [34] and can be run through a configuration file and a simple command line. Annotated polyadenylation sites were retrieved from the Polya_DB 3 and PolyASite 2.0 databases [35,36].Statistical significance of the differences between experimental and control samples was assessed by an unpaired t-test and represented using GraphPad Prism (version 9.2.0), with significance achieved at p < 0.05.In order to study the effect of the selective regulation of the two STAT1 protein isoforms on PD-L1 expression, we depleted each of them separately by using specific siRNAs (Materials and Methods) in the BRAF V600E mutant A375 melanoma cell line. As previously shown [16], the presence of PD-L1 at the cell surface (determined by FACS analysis) was inhibited by silvestrol in interferon γ (IFN-γ) treated cells (Figure 1A,B). Cells were transfected with siRNAs that selectively depleted either the STAT1α or STAT1β isoform, as shown by Western blot with an antibody detecting both protein isoforms (Figure 1C, and Figures S1C and S8). In STATα-depleted cells, IFN-γ-induced PD-L1 expression was almost entirely lost; hence, no effect of silvestrol was observed on PD-L1 levels. In contrast, in STAT1β-depleted cells, IFN-γ-induced PD-L1 expression as well as its inhibition by silvestrol were similar to that observed in control cells (that is, cells transfected with a control siRNA; Figure 1A,B). Thus, IFN-γ-induced PD-L1 expression is controlled by STAT1α, and this isoform is necessary to observe silvestrol-dependent inhibition of PD-L1 expression. This was also validated in the NRAS Q61R mutant cell line SKMEL-2 (Figure 1D and Figure S1A) and in another BRAF V600E mutant cell line WM793 (Figure 1E and Figure S1B). However, STAT1α depletion had less effects on PD-L1 expression in IFN-γ-treated WM793 cells (Figure 1E, top), as compared to the two other tested cell lines. This can be explained by the fact that WM793 cells have higher basal (without IFN-γ) expression levels of PD-L1, which is approximately three times higher than A375 (Figure S2). In the absence of IFN-γ, silvestrol did not inhibit PD-L1 expression, and no significant effect of depleting STAT1α or β isoforms was observed (Figure 1E, bottom). In addition, in WM793 cells, in spite of a higher basal level expression of PD-L1, there was still an induction of its expression upon treatment with IFN-γ (Figures S1B and S2). We therefore looked into this regulation in two other cell lines, namely MDA-MB-231 and MCF-7 (Figure S3A,B), which are breast cancer cell lines with high basal expression levels of PD-L1 (Figure S2) [37]. In these cell lines, PD-L1 expression levels were not increased by IFN-γ and were not decreased by either silvestrol treatment or STAT1 isoform depletion (Figure S3A,B). Altogether, these results indicate that the STAT1α-dependent silvestrol inhibition of PD-L1 protein expression levels is strictly manifested upon PD-L1 induction by IFN-γ, which might be of importance in the context of immunotherapy.In order to test the effect of silvestrol on the translation of STAT1α and β mRNA isoforms, we used polysome profiling in A375 cells treated with 10 nM silvestrol in the presence or not of IFN-γ. Polysome profiling uses sucrose-gradient separation of mRNAs based on their differential association with ribosomes (Figure 2A). The polysome profiles showed an overall inhibitory effect of silvestrol on translation as evidenced by a reduced polysome peak (Figure 2A). The translational status of a specific mRNA species can be inferred from its relative presence (detected by RT-qPCR) in heavier polysome fractions where it is bound to multiple ribosomes (i.e., actively translated mRNAs), lighter polysome fractions (less translated mRNAs), and subpolysomal fractions (untranslated mRNAs). The analysis of mRNAs encoding STAT1α and STAT1β was done by RT-qPCR using specific primers (Figure 2B and Table S1) and revealed that both mRNAs shifted from heavy polysome fractions to lighter polysome fractions upon silvestrol treatment, especially in IFN-γ-treated cells (Figure 2C). This indicates that silvestrol inhibits the translation of both STAT1 mRNA isoforms in IFN-γ-treated cells. Consistently, the level of both STAT1 protein isoforms, analyzed by Western blot, was decreased by silvestrol in IFN-γ-treated cells (Figure 2D, Figures S4 and S8), while there was no change in the levels of STAT1 mRNA isoforms in total cytosolic RNA (Figure S5).Since STAT1 and major immune checkpoint proteins are also expressed on immune cells, the effect of silvestrol on the translation of mRNAs encoding STAT1 and key immune checkpoint proteins, such as PD-1, CTLA-4, TIM-3 and LAG-3, was examined using polysome profiling of human peripheral blood T cells isolated from healthy donors and stimulated with IL-2 and anti-CD3/anti-CD28 antibodies. The polysome profiles showed an overall inhibitory effect of silvestrol on translation in such activated T cells, as evidenced by a reduced polysome peak (Figure 3A). Both STAT1 mRNAs isoforms shifted from heavy polysome fractions to lighter polysome fractions upon silvestrol treatment (Figure 3B), indicating that silvestrol inhibits the translation of both STAT1 mRNA isoforms in activated T cells, as observed above in melanoma cells (Figure 2C).Then, the translational status of alternatively polyadenylated mRNA isoforms encoding the key immune checkpoints PD-1, CTLA-4, TIM-3 (APA within the last exon) and LAG-3 (APA in alternative last exons) was investigated by RT-qPCR using specific primers (Figure 3C). We found that silvestrol inhibited the translation of TIM-3 mRNA isoforms #1 and #2, of LAG-3 isoform #2 (albeit less efficiently), but had no effect on translation of TIM-3 isoform #3 and all PD-1 and CTLA-4 mRNA isoforms, and had less effect on LAG-3 isoform #1 (Figure 3D). These results prompted us to extend the analysis to a few more immune-related genes. We found that silvestrol inhibited the translation of both alternatively polyadenylated mRNA isoforms transcribed from the CD137, CD27 and IDO1 genes, while it had no effect on the translation of alternatively polyadenylated mRNA isoforms transcribed from the CD40-L gene, and little effect on ADAR-1 and BTLA genes (Figure S6A,B). These results indicate that, beyond STAT1, several immune-related mRNAs (such as TIM-3, LAG-3, IDO1, CD27, CD137) are inhibited by silvestrol at the translation level. In addition, there are cases where silvestrol differentially inhibits translation of APA isoforms of the same gene (such as TIM-3 and LAG-3).Our finding that silvestrol differentially inhibited the translation of the alternatively polyadenylated isoforms of the TIM-3 and LAG-3 genes prompted us to extend this observation in a high-throughput manner. We used a 3′-seq approach that consists of targeted sequencing of the 3′-end of polyadenylated transcripts upstream of the pA tail. We carried out 3′-seq on polysomal RNA of activated T cells treated with either silvestrol or vehicle for 2 h. We found that silvestrol had a significant effect on the ratio of (Intronic Polyadenylation) IPA to (Last Exon) LE isoforms in polysomes for 1054 genes (p < 0.05; Figure 4A). For 919 genes, the polysomal IPA:LE isoform ratio was increased by silvestrol (IPA:LE up), suggesting that silvestrol had more inhibitory effect on the translation of the LE isoform compared to the IPA isoform. For 135 genes, silvestrol had more translation inhibitory effect on the IPA than the LE mRNA isoform (IPA:LE down; Figure 4A). We then concentrated our analysis on the more abundant IPA isoforms, i.e., at least 5% of the abundance of last exon “LE” isoform from the same gene (Figure 4B and Table S2). When analyzing polysomal RNAs, we identified 97 genes for which the IPA:LE ratio was increased and 59 genes for which the ratio was decreased by silvestrol. Importantly, these changes in IPA:LE ratio were only observed in polysomal RNA, but not in cytosolic RNA, for 94 (97%) of the 97 genes in the “IPA:LE up” group and for 54 (92%) of the 59 genes in the “IPA:LE down” group (Figure 4B). This indicated that, in these genes, silvestrol changes the relative translatability of the IPA and LE mRNA isoforms, without changing their relative abundance.To validate these regulation events, we focused on two genes, AHNAK (one of the 94 “IPA:LE up” candidates) and SEMA4D (one of the 54 “IPA:LE down” candidates) that have important immune functions. We selected these genes because they are relevant to T cell function. Indeed, AHNAK mediates calcium entry essential for cytolytic activity following T cell receptor (TCR) engagement of cytotoxic CD8+ T lymphocytes [38,39]. This AHNAK-dependent calcium signaling is also upregulated during CD4+ T cell activation [40,41] and a knockout of AHNAK inhibits T cell responses and the secretion of IFN-γ and IL-4 cytokines [42]. SEMA4D regulates cell proliferation by interacting with CD72 [43,44]. It downregulates the differentiation of regulatory T (Treg) cells by inhibiting Foxp3 expression [45]. In addition, antibody blockade of SEMA4D increases the efficiency of anti-PD1 and anti-CTLA4 immunotherapies [46,47].Primers were designed to quantitate alternatively polyadenylated mRNA isoforms of AHNAK and SEMA4D by RT-qPCR on polysome fractions of the Jurkat T-lymphocyte cell line treated either with silvestrol or vehicle (Figure 4C). We confirmed that silvestrol inhibited the translation of the AHNAK LE (#2) more efficiently than IPA (#1) isoform, and of the SEMA4D IPA (#1) than LE (#2) isoform (Figure 4D). Consistent with these findings, the levels of SEMA4D protein isoforms, for which antibodies were available, were differentially regulated by silvestrol in activated Jurkat T cells (Figure 4E). Whereas both SEMA4D membrane-bound and soluble protein isoforms translated from the IPA isoform were decreased in silvestrol-treated cells, the smaller protein isoform translated from the LE isoform was not decreased (Figure 4E and Figure S8).Additional candidate genes from either category, namely POU6F1, HPS3 and DENND1A (IPA:LE up) and FBXO11 and DIP2A (IPA:LE down) were also examined (Figure S7A,B). The translation of LE isoforms of “IPA:LE up” candidates and of IPA isoforms of “IPA:LE down” candidates were more strongly inhibited by silvestrol compared to the other alternatively polyadenylated mRNA isoform for each gene (Figure S7B). Altogether, these data identify genes, for which silvestrol selectively inhibits the translation of specific APA isoforms.Our findings provide several new insights into the effects of eIF4A inhibition on mRNA translation in the context of cancer.We previously demonstrated that the expression of PD-L1 in cancer cells was decreased by silvestrol, but the effect was indirect and mediated by translation regulation of STAT1, a transcriptional regulator of PD-L1 [16]. In fact, two protein isoforms of STAT1 are encoded from two alternatively polyadenylated mRNA isoforms. We extended the conclusions from these findings by showing that the silvestrol-mediated inhibition of PD-L1 expression is dependent on STAT1α, but not STAT1β. This illustrates the relevance of alternative polyadenylation in generating protein isoforms with different functions. Beyond STAT1, there are a few examples illustrating the functional relevance of alternative polyadenylation in T cells. The T-cell surface glycoprotein CD5 is regulated by alternative polyadenylation during human T lymphocyte activation. Preferential 3′UTR shortening of CD5 mRNA upon T cell activation, due to proximal pA site usage, leads to higher CD5 expression and determines cellular outcomes of survival or apoptosis [48]. Another example of regulation of alternatively polyadenylated transcripts following T cell activation relates to CELF1/CUGBP1 target transcripts. A number of CELF1 regulated transcripts involved in cell division are generated with shorter 3′ ends to promote the exclusion of CELF1 binding sites, thereby upregulating their expression following stimulation of T cells [49].Genome-wide studies showed that T-cell activation is associated with a selective usage of proximal pA sites, leading to a preferential increase in the expression of mRNAs with shorter 3′UTRs [50,51]. However, in most genes, this preferential shortening of the 3′UTR is not associated with increased protein production [50,52]. Here, by combining RNA analysis (through 3′-seq or RT-qPCR) and polysome profiling, we provide evidence that silvestrol differentially inhibits the translation of mRNAs with different 3′-ends produced through alternative polyadenylation. The main determinant of an mRNA for its dependence on the eIF4A RNA helicase is the level of stable secondary structures [53] or more complex G-quadruplex structures [54] within the 5′ UTR of mRNAs. It is possible that the 3′UTR of an mRNA influences 5′UTR-based mechanisms when the UTRs are in close proximity and that this makes an mRNA sensitive to silvestrol inhibition for certain 5′UTR-3′UTR combinations. Such proximity between the 5′UTR and 3′UTR has, for instance, been observed by electron microscopy on the membrane of the endoplasmic reticulum, where the great majority of polysomes have a circular shape [55,56]. It is also possible that the 3′UTR is involved in mRNA localization in specific subcellular regions that are more prone to silvestrol inhibition. It is very well admitted that 3′UTRs influence mRNA localization to very specialized cellular sites such as dendrites or synapses in neuronal cells [57,58]. mRNA localization is dependent on the repertoire of 3′UTR-bound RNA binding proteins. For instance, the TIS11B RNA binding protein interacts with specific 3′UTR-containing mRNAs to form membrane-less organelles called TIS granules intertwined with the endoplasmic reticulum [58]. Interestingly, several membrane protein-encoding mRNAs, including PD-L1, predominantly localize to TIS granules [58]. In addition, 3′UTRs are determinants of mRNA localization in stress granules [59,60]. It has been shown that inhibitory immune checkpoint mRNAs (including TIM-3 and LAG-3) require microtubule-dependent stress granule dynamics [61]. All these 3′UTR-dependent mechanisms should be carefully tested in the context of the 3′UTR-dependent inhibition of translation by silvestrol.Finally, silvestrol is shown here for the first time to inhibit the translation of mRNAs encoding key immune checkpoint inhibitors (e.g., TIM-3 and LAG-3) in T cells. Considering the promising results achieved with inhibitors of LAG-3 [62] and TIM-3 [63], and the fact that eIF4A inhibitors are currently tested in clinical trials, it will be interesting to test combinations of eIF4A inhibitors with either TIM-3 or LAG-3 inhibitors. In this context, it will be essential to consider the 3′UTR-dependent inhibition of translation by eIF4A inhibitors to better understand the potential beneficial effect of these treatment combinations.Our findings reveal that inhibiting the eIF4A RNA helicase component of the eIF4F translation initiation complex with silvestrol differentially inhibits alternatively polyadenylated mRNA isoforms of certain immune-related genes like TIM-3, LAG-3, AHNAK and SEMA4D, among others. They have important implications for treatment of human cancers with eIF4A inhibitors while some of them are entering into clinical trials. Investigating the impact of eIF4A targeting on the translation of alternatively polyadenylated mRNA isoforms of the same gene may contribute to the understanding of anti-cancer activities of eIF4A inhibitors.The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers14051177/s1, Figure S1: (A) PD-L1 was visualized by flow cytometry in SK-MEL-2 cells treated with IFN-γ (100 ng/mL) and silvestrol (10 nM or 30 nM). (B) PD-L1 was visualized by flow cytometry in WM793 cells treated with or without IFN-γ (100 ng/mL) and silvestrol (10 nM or 30 nM). (C) Western blot analysis to look into the effect of silencing either STAT1 isoform in A375 cells treated with or without IFN-γ for 24 h and silvestrol (10 nM or 30 nM) for 24 h. Quantification of STAT1 expression was performed by calculating the relative densities normalized to GAPDH levels. One representative blot from three independent experiments is shown; Figure S2: PD-L1 was visualized by flow cytometry in A375, WM793, MCF-7 and MDA-MB-231 cells treated without or with IFN-γ (100 ng/mL). Bottom: PD-L1 mean fluorescence intensity (MFI) quantification. The data are presented as the mean ± s.e.m. (n = 3 independent experiments). p-values were calculated using a two-tailed unpaired t-test; Figure S3: Left: PD-L1 was visualized by flow cytometry in (A) MCF-7 and (B) MDA-MB-231 cell lines treated with or without IFN-γ (100 ng/mL) and silvestrol (10 nM or 30 nM). Right: PD-L1 mean fluorescence intensity (MFI) quantification. The data are presented as the mean ± s.e.m. (n = 3 independent experiments). p-values were calculated using a two-tailed unpaired t-test. Bottom, Western blot analysis of the indicated proteins; Figure S4: Western blot analysis to look into the expression of STAT1 APA isoforms in SK-MEL-2 and WM793 cells treated with IFN-γ for 24 h and silvestrol (10 nM or 30 nM) for 24 h. Quantification of STAT1 expression was performed by calculating the relative densities normalized to GAPDH levels. One representative blot from three independent experiments is shown; Figure S5: Relative mRNA expression of total and APA isoforms of STAT1 and of TBP (control) in total (non-fractionated) lysates of A375 (n = 3). A375 cells were treated without or with IFN-γ for 24 h and silvestrol (10 nM or 30 nM) for 24 h; Figure S6: (A) Primer design for immune-related gene mRNA isoforms. (B) Percentage of transcripts for each APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation was quantified by RT-qPCR (n = 3). p-values were calculated using two-tailed unpaired t-test (* p ≤ 0.05); Figure S7: (A) Primer design for immune-related gene mRNA isoforms of IPA:LE UP (red) and IPA:LE DOWN (blue) candidates. (B) Percentage of transcripts for each APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation was quantified by RT-qPCR in Jurkat cells (n = 3). p-values were calculated using two-tailed unpaired t-test (* p ≤ 0.05); Figure S8: Uncropped Western blot images; Figure S8: Uncropped Western Blot original images; Table S1: Primer sequences used for RT-qPCR analysis of all mRNAs investigated in this paper; Table S2: List of genes with upregulated and downregulated IPA:LE isoform ratio, as identified from 3′ Sequencing analysis in heavy polysomes and whole cytosol. Columns A–C, coordinates of IPA peaks (hg19). Column D, IPA peak number in our analysis. Columns E–F and I, gene coordinates. Column G, gene symbol (please note that some genes contain several regulated IPA peaks). Column H, RefSeq information on coding (NM) or noncoding (NR) status of gene transcripts. Columns N–Y: Number of reads in each sample either in IPA peak (IPA) or in the gene’s last exon (LE). Biological replicates are indicated (n1 to n3). Silv, Silvestrol. HP, heavy polysomes. CYTO, whole cytosol. Column Z, abundance of IPA peak relative to last exon (in %). Columns AC and AD, hg19 coordinates of annotated polyA site(s) in published databases [35,36] overlapping with the regulated IPA peak.Conceptualization, B.B., R.G., C.R. and S.V.; Methodology, B.B., M.C., M.D., C.M.L., A.C. and R.G.; Validation, B.B. and R.G.; Formal Analysis, all; Investigation, all; Writing—Original Draft Preparation, B.B., C.R. and S.V.; Writing—Review and Editing, all; Visualization, B.B., C.R. and S.V.; Supervision, C.R. and S.V.; Funding Acquisition, R.G., C.R. and S.V. All authors have read and agreed to the published version of the manuscript.This research was funded by CNRS, Institut Curie, Ligue contre le Cancer Équipe labellisée to S.V., Collectif Ensemble contre le mélanome, Association Vaincre le mélanome to C.R. and INSERM and INCA-PRTK to S.V. and C.R. The APC was funded by Ligue contre le Cancer.Not applicable.Not applicable.The 3′-seq data presented in this study are openly available in the Gene Expression Omnibus repository (GEO) under accession number GSE193966 (token: cdkviuggdjoldin). Computer Code and Software: The complete bioinformatics pipeline 3′-SMART can be freely downloaded at GitHub (https://github.com/InstitutCurie/3-SMART, accessed on 19 January 2022).We thank the Institut Curie Next Generation Sequencing (Sylvain Baulande) platform for high-throughput sequencing.C.R. is an occasional consultant for Bristol Myers Squibb, MSD, Novartis, Sanofi, AstraZeneca, Pfizer, Roche and Pierre Fabre. C.R. and S.V. are scientific founders of Ribonexus. B.B., R.G., C.L. and M.D. declare no conflict of interests.Functional importance of APA-generated STAT1 protein isoforms on PD-L1 gene expression. (A) PD-L1 was visualized by flow cytometry in A375 melanoma cells treated with IFN-γ (100 ng/mL) and silvestrol (10 nM or 30 nM); (B) PD-L1 mean fluorescence intensity (MFI) quantification in A375. The data are presented as the mean ± s.e.m. (n = 3 independent experiments). p-values were calculated using two-tailed unpaired t-test. Statistical analyses were performed for all data, p-values are indicated exclusively in the case of statistical significance; (* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001) (C) Western blot analysis of the indicated proteins in A375. Quantification of STAT1 expression was performed by calculating the relative densities normalized to GAPDH levels (uncropped western blot original images see Figure S8). (D) same as (B) in SKMEL-2 (E) same as (B) in WM793 with (top) or without (bottom) IFN-γ.Both STAT1 mRNA isoforms are regulated at the translational level by eIF4A inhibition. (A) Polysome profiles of A375 cells treated with IFN-γ for 24 h and silvestrol (10 nM) for 2 h. One representative profile from three independent experiments is shown; (B) primer design for RT-qPCR detection of STAT1 mRNA isoforms; (C) percentage of transcripts for each STAT1 APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation was quantified by RT-qPCR (n = 3). p-values were calculated using two-tailed unpaired t-test (* p ≤ 0.05); (D) Western blot analysis to look into the expression of STAT1 APA isoforms in A375 cells treated with IFN-γ for 24 h and silvestrol (10 nM or 30 nM) for 24 h. Quantification of STAT1 expression was performed by calculating the relative densities normalized to GAPDH levels. One representative blot from three independent experiments is shown (uncropped western blot original images see Figure S8).eIF4A inhibition regulates the translation of mRNAs encoding key immune checkpoint proteins such as TIM-3 and LAG-3, but not PD-1 and CTLA-4, in activated T cells. (A) polysome profiles of T cells stimulated for 72 h and treated with silvestrol (10 nM or 30 nM) for 2 h. One representative profile from three independent experiments is shown. (B) Percentage of transcripts for each STAT1 APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation of T cells was quantified by RT-qPCR (n = 3). p-values were calculated using two-tailed unpaired t-test (* p ≤ 0.05); (C) primer design for immune checkpoint gene mRNA isoforms; (D) percentage of transcripts for each APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation was quantified by RT-qPCR (n = 3). An asterisk (*) was inserted for each fraction to indicate a statistically significant difference (p < 0.05).eIF4A inhibition regulates the translation of several immune-related alternatively polyadenylated mRNAs. (A) Regulation of intronic polyA (IPA) versus last exon (LE) isoforms in heavy polysome fractions following treatment with silvestrol (10 nM) for 2 h in CD8+ T cells from healthy donors. IPA sites that are significantly upregulated (IPA:LE up) or downregulated (IPA:LE down) by silvestrol relative to LE (p-adj < 0.1) are shown in red and blue, respectively; (B) number of genes with either up- or downregulation of IPA:LE isoform ratio by silvestrol in heavy polysome fractions or whole cytosol, as indicated; (C) primer design for candidate immune-related gene mRNA isoforms; (D) percentage of transcripts for each APA isoform in each polysomal fraction obtained by sucrose-gradient ultracentrifugation was quantified by RT-qPCR in Jurkat cells (n = 3). p-values were calculated using two-tailed unpaired t-test (* p ≤ 0.05); (E) Western blot analysis to look into the expression of SEMA4D APA isoforms in Jurkat cells without or with stimulation for 72 h including silvestrol (10 nM) treatment for 24 h. Quantification of SEMA4D expression was performed by calculating the relative densities normalized to GAPDH levels. One representative blot from three independent experiments is shown (uncropped western blot original images see Figure S8).Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Med-MDPI/cancers_2/cancers-14-05-01178.txt

@@ -0,0 +1,3 @@
+Previous research describes the issues AYA cancer patients may face when it comes to maintaining social relationships after their diagnosis. Related issues included mutual misconceptions and a lack of understanding of the impact of cancer. The Dutch AYA ‘Young & Cancer’ Care Network co-created the mobile application ‘AYA Match’ to provide support on this matter. Co-creation, in which the target population is directly involved, appears to be an effective way to establish an intervention that applies to their needs. The aim of this study was to describe the cocreational process, characteristics of AYA Match users and their expectations of the app. The application could be useful for a wider audience in the future, such as older cancer patients or individuals dealing with other diseases.Adolescent and young adult (AYA) cancer patients report a need for support to stay in contact with loved ones after diagnosis. In response to this the Dutch AYA ‘Young & Cancer’ Care Network co-created the mobile application ‘AYA Match’. This study describes the cocreational process, the characteristics of the users and their expectations regarding the app. 121 AYA cancer patients and 37 loved ones completed a questionnaire. 68.6% of the loved ones reported ‘staying in contact’ and ‘finding out about the needs and wishes of ‘their AYA’ during this time’ as the main reasons for downloading the application. 41.1% of the AYA cancer patients expected the app to help them communicate to their loved ones what they do or don’t want and need. 60% of the loved ones indicated that they would like to use the application to offer help to ‘their AYA’ with their daily tasks. Patients and their loved ones have similar expectations when it comes to ‘normalizing’ contact, increasing empathy and mutual understanding about needs and emotions. The AYA Match app could be an adequate answer to the issues experienced regarding contact, support and mutual understanding.In the Netherlands, each year approximately 3900 adolescents and young adults (AYAs), aged 18 to 39 years, are diagnosed with cancer for the first time [1]. They constitute a specific patient group in oncology, considering their unique medical and psychosocial characteristics. AYAs with cancer differ from children and older adults with regard to the broad spectrum of cancer diagnoses, tumor biology and physiology, pharmacology and genetic variance with respect to cancer susceptibility and treatment [2]. Even though AYAs share many cancer-related challenges with pediatric and older adult cancer patients, including short-term (such as hair loss, pain) and long-term effects of treatment (such as fatigue, cognitive problems, increased risk of cardiovascular diseases), these issues may have a larger impact and different effects on AYA cancer patients. Adolescence and young adulthood are challenging life phases of physical, emotional, cognitive and social maturation. Dealing with a life-threatening disease whilst trying to achieve developmental milestones can have a negative impact on the health-related quality of life of AYA cancer patients [3,4,5].Adolescents and young adults are dealing with important and future-shaping matters such as completing education, becoming financially and socially independent, developing their own identity [4,5], exploring sexuality, building long-term relationships and deciding on whether they want to become a parent [3,5,6]. The sudden disruption of their (social) life, possible changes in physical appearance, confrontation with mortality, increased temporary dependence on informal caregivers like partner or parents, the risk of losing reproductive capacity and other health-related concerns caused by their cancer diagnosis or treatment disturb or even disrupt the ‘normal’ development of these young people [5,6,7,8]. AYAs diagnosed with cancer report lower social functioning levels compared to their healthy peers even two years after diagnosis [8]. They also report more social functioning problems compared with older cancer patients [9,10]. A qualitative study by Breuer and colleagues (2017) stated that AYA cancer patients indicate their family, friends and partners as the most important people to receive social support from [11]. Breuer and colleagues indicate that, besides helpful support, AYA cancer patients also experience negative support, such as people being overly careful instead of treating them ‘normal’ or understating the severity of their illness. They frequently report difficulties when it comes to establishing and maintaining social relationships, due to long-term effects of treatment or having the feeling of being different from their healthy peers [8]. This may result in friends decreasing or even completely breaking off contact [11].The vast majority of adolescents and young adults are frequent users of digital applications. In 2018, 97% of the Dutch population had access to the internet and 90% of those aged 16 to 74 used the internet daily [12]. This increases the relevance of implementing digital interventions. A study by Allison and colleagues (2021) indicates the value of online platforms where individuals with cancer are provided with the opportunity to discuss experiences, show emotions and feel understood [13].Dutch AYA cancer patients in particular highly value the use of online communities focused on peer support, when it comes to recognition and receiving advice and support regarding problems. In a study by Kaal and colleagues (2017) on the usefulness of the Dutch online community for AYA cancer patients, almost half of the participants reported feeling less lonely due to being part of an online peer community [14].Despite the growing evidence of the usefulness of e-health interventions facilitating social and peer support, not many of the available interventions focus on communication and ‘normalizing’ contact with friends and family during illness. Most digital interventions targeting AYAs with cancer, e.g., web-based or mobile applications, focus mainly on physical health and self-management of side and late effects [15].In 2016, a group of Dutch AYA cancer patients and their loved ones (e.g., friends, family, colleagues) actively put forward the need for support to stay in contact with each other after the cancer diagnosis (www.ayazorgnetwerk.nl, accessed on 27 October 2021). Related issues included mutual misconceptions and the lack of understanding of the impact of cancer, as well as trouble maintaining social relationships. However, the main issue seemed to be that all conversations between AYAs and their loved ones came back to the illness of the AYA. ‘Let me know if there’s anything I can do for you’, often remains unanswered and therefore turns out to be more dismantling than helpful in the relationship. This led to disconnecting AYAs from who they are and what matters to them as a person, beyond their cancer diagnosis. The call of AYAs and their loved ones to the Dutch AYA ‘Young and Cancer’ Care Network, regarding the need for support, led to the co-creational process of developing a digital intervention: the AYA Match app [16]. The process started with an extensive exploratory study conducted by Engelen (2017) assigned to the Dutch Federation of Cancer Patients Organizations examining the needs of AYA patients and their loved ones regarding social contact and mutual understanding after a cancer diagnosis [16].This article describes (1) the co-creational process of the development of the AYA Match app and its features; (2) the characteristics of the AYA Match app users, and (3) users’ expectations of the AYA Match app.Through the entire co-creational process of developing the ‘AYA Match’ app, the most important precondition was the involvement of the target users: AYA cancer patients and their loved ones. “Co-creation is a process known as bringing different parties together, which produces a blend of ideas leading to a mutually valued outcome” [17].Within the co-creational process, a pitch meeting, a ‘canvas session’, a ‘paper prototype session’, a ‘feedback session’ and a ‘test phase’ were conducted (Figure 1). All sessions and phases were facilitated by the digital developers of the application.In all phases different AYA cancer patients and/or loved ones were actively involved and considered primary stakeholders throughout the co-creational process. The majority of them had been active within different project groups and think tanks of the Dutch AYA ‘Young & Cancer’ Care Network for a longer period and were therefore considered ‘experts by experience’. Their variety in diagnosis, age, sex and educational level, as well as including new AYAs cancer patients and loved ones during each phase of the cocreational process, contributed to the overall representability.Five digital development companies were approached by the Dutch AYA ‘Young & Cancer’ Care Network to share their ideas on a digital tool (serious game, mobile application or other). This tool had to meet the needs of AYAs with cancer and their loved ones regarding social support. The tool had to be easy to use and accessible. The ultimate goal was that the AYA cancer patient felt more connected to and understood by friends, family and other acquaintances. The companies pitched their idea in person to several AYA cancer patients and representatives of the Dutch AYA ‘Young & Cancer’ Care Network.The winning pitch presented the idea of a mobile application called ‘AYA Match’. This idea was chosen by the attending AYA cancer patients and their loved ones. The app was developed by using an iterative approach, meaning repeated testing of a concept by the end users, resulting in an application entirely built around their wishes and needs.During the canvas session two AYA cancer patients and a loved one were motivated to discuss and brainstorm through a two hour face-to-face focus group meeting, in order to ascertain which components were needed within the application to reach the desired outcome.For the paper prototype session seven AYA cancer patients and their loved ones were invited to interactively try out the concept during a face-to-face focus group meeting, using printed cards containing the different screens of the potential application (Figure 2). By actually trying out the concept instead of just presenting it, honest and specific feedback was elicited. During this session oral feedback was obtained on the first nondigital version of the prototype.During the feedback session the virtual design of the application was presented on a big screen to five AYA cancer patients and their loved ones, who were able to orally provide the designers with feedback on the visual elements before the start of the test phase.This latter period covered six weeks of testing the digital prototype of the application by five AYA cancer patients and their loved ones. The testing took place at their own homes in their own time. The goal was to find out whether the application stimulated contact between both parties as was aimed for. During this phase, they shared their thoughts on and findings of the different features and screens with the designers via WhatsApp (https://www.whatsapp.com, accessed on 4 February 2017) and an online survey via SurveyMonkey (https://www.surveymonkey.com, accessed on 4 February 2017). The online survey consisted of 13 questions on the usability of the app (such as: ‘What do you find best and worst about the app?’; ‘Is it easy to invite loved ones?’; ‘Are the different features user friendly?’), possible improvement (such as: ‘What changes would you suggest should be made?’; ‘Is the content suitable?’) and whether it served its initial purpose (such as: ‘Do you think the app initiates and stimulates contact?’; ‘Does the app help to discuss heavy topics?’; ‘Does the app cause activities to happen?’).Once the final concept (titled: AYA Match) was launched as an openly accessible and free mobile application, all users (AYA cancer patients and their loved ones) were invited to participate in a descriptive quantitative questionnaire study.Study eligibility criteria included AYAs diagnosed with any type of cancer for the first time between the ages of 18 and 39 years or being a loved one (no age limit) of an AYA cancer patient, including parents, partners, friends and colleagues for example. Participants had to be able to read and understand Dutch.No targeted recruitment regarding the questionnaire study was conducted. All users of the AYA Match app were considered potential participants and invited to take part in the study via the onboarding screens of the app. The app was promoted among different AYA stakeholders in a number of ways. First, the AYA Match app was launched at the Dutch annual AYA ‘Young & Cancer’ congress in March 2018, where attending AYAs were encouraged to download the app. In addition, promotion material, such as flyers, social media posts and a video, was disseminated through patient organizations and health care professionals during the following months. Everyone who downloaded the app was invited to take part in the study and to complete the baseline questionnaire online before they started using the app. If a user decided not to take part in the study, there were no consequences for their further usage of the app. Those who decided to participate were automatically forwarded to the online informed consent form and baseline questionnaire. Two versions of the baseline questionnaire were created: one for AYA cancer patients and one for their loved ones.Questionnaires were programmed within the online survey platform ‘SurveyMonkey’.The questionnaire for the AYA cancer patients consisted of, among others, questions on sociodemographic and clinical data, an item on their intrinsic motivation to download the app and items on their expectations of usage (such as ‘Who (e.g., family, friends, colleagues) would you like to invite to use the app with?’; ‘What are you hoping the app will help you with?’).The questionnaire for the loved ones consisted of, among others, questions on sociodemographic data, clinical data of ‘their AYA’, an item on their intrinsic motivation to download the app and an item on their expectations of usage (such as: ‘What is the reason you downloaded the app?’, ‘What are you hoping the app will help you and ‘your AYA’ with?’).Analyses were performed using SPSS statistical software (version 26, Chicago, IL, USA). Descriptive statistics and frequencies concerning sociodemographic data, clinical data, and user expectations were calculated.The winning pitch presented the idea of a mobile application offering support on staying in contact after a cancer diagnosis: AYA Match.During the canvas session, the purpose of the application was defined and described as follows:
+          ‘Create a tool to spark contact between both parties:The app (re)connects AYA cancer patients with their loved ones (e.g., family, friends, colleagues). In a casual, unforced way.’Questions that were difficult to ask become less heavy. ‘Let me know if there’s anything I can do for you’ will no longer remain unanswered.An ‘appetizer’ to spark contact, fitting the moment.‘Create a tool to spark contact between both parties:The app (re)connects AYA cancer patients with their loved ones (e.g., family, friends, colleagues). In a casual, unforced way.’Questions that were difficult to ask become less heavy. ‘Let me know if there’s anything I can do for you’ will no longer remain unanswered.An ‘appetizer’ to spark contact, fitting the moment.Feedback obtained during the canvas session included comments from loved ones such as: 
+          “Who am I? Is it up to me to bring up this topic? I have no idea if they have someone to talk to?”, “I’m not sure what they want or don’t want. I find it complicated to ask. I’m not sure if it’s awkward to ask about their needs…” and “I would love to help, but I’m not sure how to do that. I said: let me know if there’s anything I can do, but haven’t heard from them since…”.“Who am I? Is it up to me to bring up this topic? I have no idea if they have someone to talk to?”, “I’m not sure what they want or don’t want. I find it complicated to ask. I’m not sure if it’s awkward to ask about their needs…” and “I would love to help, but I’m not sure how to do that. I said: let me know if there’s anything I can do, but haven’t heard from them since…”.This feedback was used by the digital developers to create a paper prototype to try out and discuss with the AYA cancer patients and their loved ones in the next phase.The paper prototype was used to try out and prioritize possible features within the app, resulting in the further development of the two key elements:‘Rules to play’ is meant to inform loved ones on the patient’s preferences regarding communication around their illness. The patient can ‘like’, ‘dislike’ or ‘pass’ on the different statements. The loved ones can view the patient’s preferences;The ‘activity cards’ give both parties the opportunity to match on a variety of fun things to do, and to ask for (patient) or offer (loved one) help. The activity cards are followed up by a feature where the patient can choose a match with the person of preference and continue to set a date in the built-in chat feature.‘Rules to play’ is meant to inform loved ones on the patient’s preferences regarding communication around their illness. The patient can ‘like’, ‘dislike’ or ‘pass’ on the different statements. The loved ones can view the patient’s preferences;The ‘activity cards’ give both parties the opportunity to match on a variety of fun things to do, and to ask for (patient) or offer (loved one) help. The activity cards are followed up by a feature where the patient can choose a match with the person of preference and continue to set a date in the built-in chat feature.These two elements mainly focus on the practical aspect of maintaining contact: ‘how can I ask for/offer help’, ‘what are topics the patient does or does not want to talk about’ and ‘how can I communicate my needs’. More game-oriented elements were suggested, but not preferred by the involved AYA cancer patients and loved ones.Feedback obtained during the paper prototype session included comments such as: “if I would’ve had access to an app like this during my treatment, I am absolutely sure I wouldn’t have lost as many friends as I did”, confirming the usefulness of the concept. In addition, ideas for refinement were listed, such as “we need to add more humor”.The feedback session was used to present the concept of the different screens and features of the first prototype of the AYA Match app. The visual aspect of the app was developed and presented step-by-step, as shown in Figure 3. The visual elements (e.g., colors, typography, navigation bar and buttons) were designed and adjusted based on the preferences of the patients and loved ones present.During the six-week test phase AYA patients and their loved ones were asked to test the prototype version of the application. The application developer IJsfontein obtained their final feedback through SurveyMonkey and WhatsApp.The reports of the testers included positive feedback on easy usage (quote 1, 12, 15) and efficiency (quote 5, 7, 9, 10, 12, 13, 14, 18) (Table 1).Based on the reported outcomes, final adjustments were made such as the option to add a profile picture (quote 3), structured onboarding (quote 2, 6, 8), a built-in chat function (quote 4), the possibility to skip propositions on certain topics (quote 9, 11), creating an overview of matches per person (quote 16) and activating push notifications to increase engagement (quote 17).The final phase was kicked off with the official launch of the application, which enabled anyone interested to download and start using AYA Match. The application ended up with two main features, as described in Section 3.1.2 and shown in Figure 4.Participants were not personally invited and as a result no detailed information of respondents versus non-respondents was available.In total 141 participants took part in the study. However, three did not complete the full questionnaire and 17 were outside the AYA age range at time of diagnosis, resulting in 121 eligible participants. Completing the questionnaire took approximately eleven minutes. Table 2 displays the sociodemographic and clinical characteristics of the participants. Among them were 100 female and 21 male participants. Most of the participants (n = 114) were residents of the Netherlands, versus seven Belgian participants. Mean age at baseline was 28.8 ± 4.3 years old. Mean age at diagnosis was 27.2 ± 4.3 years old. Thirty-eight percent (n = 46) of participants were diagnosed with breast cancer. Forty-nine participants (40.5%) were within their first year after diagnosis, 52.1% were between one and five years after their diagnosis (n = 63) and nine were between six to ten years post diagnosis.In total, 74 loved ones participated in the study. However, seven of the participants completed the baseline questionnaire twice, resulting in a total of 67 participants. Of them, 30 were excluded because the patient who invited them to use the app was outside the AYA age range (n = 37). Completing the questionnaire took approximately nine minutes. Sociodemographic and clinical data are reported in Table 2.Among the eligible participants were 27 females and 10 males. 33 participants were residents of the Netherlands, versus 4 Belgian participants. Mean age at baseline was 39.9 ± 14.7 years old. Almost half of them (45.9%) were friends, others were partners, family members and colleagues. Fifty-one percent of ‘their AYAs’ were on active treatment at baseline.Table 3 describes the expectations of the remaining participants (n = 107, AYA cancer patients, n = 35 loved ones) of the AYA Match app. Almost 25 percent of patients reported ‘staying in contact’ as an important reason for downloading the application, as did 68.6% of the loved ones.60.0% of the loved ones indicated that they would like to use the application to offer help to their AYA with their daily tasks. 71.0% of the patients would like to use the application with their close friends, but they also reported considering inviting their parents (41.1%) and their partner (35.5%).The primary aim of this study was to describe the co-creational process of developing the AYA Match app and report the characteristics of its users and their expectations of the app. Over 40% of AYA cancer patients included in this study indicated the hope that using AYA Match would help them clarify to loved ones what they want and need. Over 65% of responding loved ones stated they expected the app would help them understand what they could do to support ‘their AYA’. In addition, 43% reported the app might help them gain a better understanding of the situation ‘their AYA’ was in and how they are feeling. This study has shown that AYA cancer patients and loved ones have similar expectations of the app when it comes to contact and mutual understanding after a cancer diagnosis. Due to the process of co-creation, the needs of both parties have been integrated into a digital intervention. As far as known to the authors, there aren’t many comparable apps described in literature. AYA Match intervenes on communication with friends and family during treatment, while most digital interventions focus on lifestyle, symptom registration, communication with doctors and other cancer patients [15].It is notable that the vast majority of respondents in this study were female, which is likely related to the high number of breast cancer patients. Most of the included AYA cancer patients were between 18–29 years of age at diagnosis, and over half of the participants were in the phase of active treatment. This indicates that AYAs in active treatment might experience a bigger need for support in communication and contact with the people around them compared to those who already completed treatment. Almost all AYA cancer patients and loved ones finished higher level education. A study by Jansen and colleagues (2015) demonstrates that being younger, higher educated and under active treatment is related to a positive outlook on the use of eHealth [18]. Caroll and colleagues (2017) reported that patients of older age, male gender and who have a lower educational level (secondary vocational education or lower) indicated a reduced probability of using a health app [19].Co-creation can be seen as an important aspect of a participatory design (PD). This is a collective and creative process, which includes five phases: exploration, creation, testing, implementing and refreshing. “In PD, one of the guiding principles is that every relevant stakeholder can and should be involved in the design process.” The Dutch AYA ‘Young & Cancer’ Care Network developed a method for successfully involving AYA cancer patients in improving age-specific healthcare and tools, such as the AYA Match app [20].A study by Bjerkan and colleagues (2014) endorses the value of a flat structure, a method in which patients’ expectations and their knowledge is directly communicated to system developers. The flat structure was used while creating the AYA Match app. By creating a confidential environment wherein all stakeholders are represented, individual needs, expectations and new ideas are expressed and used to the utmost potential [21].This indicates that co-creation can be a useful method of converting the patients’ needs and wishes into a self-management intervention. The similarity between the input of patients and loved ones collected during the co-creational process and the expectations of the actual users of the intervention, underline the usefulness of co-creation in the development of the AYA Match app.A study by McNeil (2019) indicates that social support from family and peers is of significant importance for AYA cancer patients. This support, however, is a changing concept over time and may thus differ throughout one’s treatment period [22]. Breuer and colleagues (2019) reported that the relationships with loved ones are affecting how the AYA cancer patient feels, by providing support and listening to them. Undertaking activities together, as the AYA Match app stimulates, was experienced as distracting from living with a disease and perceived as pleasant. The participants in the study indicated that it was helpful if loved ones were available and reachable throughout treatment [11]. These studies substantiate the importance of the AYA Match app. By indicating one’s needs for support at individual moments in time, this application responds to the changing aspect of support and facilitates continuous accessible contact with loved ones.It should be taken into account that the co-creational process described in this paper was conducted in a small sample of AYA cancer patients and their loved ones. In addition, informed consent was not sought during the co-creational process. As a result it was not possible to share their sociodemographical data. This may make generalization more difficult. It could be challenging to recruit many diverse participants, as was experienced during recruitment. However, research has shown that it is not necessary to have a large group in order to properly implement a participatory design [20]. The AYAs and loved ones who contributed in the co-creational process have been involved in their patient organization for some time and are thus familiar with the needs of other patients and their loved ones, besides their own experiences. A larger group may be recruited for the follow-up research, but for this developmental study, the contribution of these participants was estimated to be very valuable.During the developmental phase of the AYA Match app, the AYAs involved in the co-creational process were almost all in the post-treatment phase. The problem of recruiting patients undergoing active treatment is widely acknowledged in literature but under-researched [23,24]. In addition to factors identified by Van den Brink et al. (2020) such as low disease incidence, low project awareness, and high patient burdens (e.g., time commitment), we propose additional factors specific for our target group of AYAs, namely (1) low physiological and/or emotional wellbeing as a result of the disease and/or treatment, (2) high (perceived) risks of meeting up in groups (e.g., due to compromised immune systems), and (3) high avoidance of emotional distress (e.g., reflecting on the social-psychological impact of one’s disease) [24]. While a deep dive into these factors was out of scope for our project, this stands as a key recommendation for further research.Our survey has been carried out on a very diverse population, given the broad age range and cancer types of the participating AYAs. This increases the external validity of the research findings and allows for possible generalization. Yet generalizing the results of this study should be done with caution, given its limitations. Those who participated in this research included a large number of female cancer patients and many patients were in active treatment at the time of downloading the application. The need to improve social support and contact with friends and family is not only important during treatment: patients are in need of interventions to decrease the lack of social support post-treatment as well. Research indicates a significant drop in the quantity of social support from loved ones over the first-year post-treatment [25]. In addition, it has been found that AYA cancer patients in particular score significantly lower on social functioning up to two years after diagnosis compared to their healthy peers, and one-third of them remain at risk of poor social functioning [8]. Therefore, an intervention that helps harmonizing the social relationships of the AYA during treatment might prevent social isolation later on.This study was conducted using self-reporting, which is limited by recall bias. Also, self-reports have a risk of social-desirable answers. It could be difficult to admit that one wished for more social support or that one does not want it to be inferior to the support they did receive. As the questionnaire focused on user expectations and therefore was completed once, the answers might be influenced by one’s mood at that time.This study adds to the current knowledge by identifying the needs and wishes of AYA cancer patients regarding social support and their outlook on using applications to help them receive this. Social support coming from friends and family during times of treatment and rehabilitation contribute to the extent of how well patients adjust to their new situation [26]. However, staying in contact and leveling with family and friends can be difficult after a cancer diagnosis. Especially young cancer patients are at risk of poor social functioning and therefore in need of interventions stimulating social support [8]. A review by Tremolada and colleagues (2016) demonstrates that female AYA cancer patients prefer to ask directly for support from family and friends when it comes to solving problems. Male AYA cancer patients favor more indirect strategies to receive support, for example initiating pleasant, physical activities [27]. In addition, even though the application has been developed in co-creation, the design may not be attractive to men. Future research should be conducted to determine potential gender differences in use of the application that could inform future refinement of the application to make it attractive for both sexes.The AYA Match app could potentially be considered an adequate answer to the needs of AYA cancer patients and their loved ones, as it focuses on initiating and maintaining contact. Both the possibility to seek support among loved ones and the encouragement of undertaking activities together are facilitated within the intervention, serving both the preferences of males and females as described by Tremolada [27]. Therefore, extended research is being conducted to measure the effect of using AYA Match on social functioning and perceived social support during treatment, as well as the impact of a cancer diagnosis on the social life of AYA patients.Parallel to the continuous research, partnerships with fellow organizations are established to create sustainable support for further development of the application and make it available to a larger number of people. Several parties have shown their interest in the AYA Match app, including healthcare providers of pediatric- and older adult cancer patients, as well as (international) patient organizations of other diseases and humanitarian organizations. This interest has been endorsed by the 30 loved ones who were using the app, but were excluded in this study due to ‘their AYA’ being outside the AYA age range.In extend to the search for sustainable support, the AYA Match app is being re-explored and possibly renamed to make it more appealing to patients outside the AYA age range.In conclusion, the results as described in this study demonstrate that the application could be an adequate answer to the issues AYA cancer patients and their loved ones experience when it comes to contact and mutual understanding after a cancer diagnosis. Co-creation, in which the target population is directly involved in the process of developing an intervention, appears to be an effective way to establish an intervention which applies to their needs. By sharing our knowledge and possibly including other patient populations, such as older cancer patients or individuals dealing with other diseases, the application may become available for a wider audience in the future.The following supporting information can be downloaded at https://youtu.be/78EAqBUxHAM (accessed on 5 December 2018), Video S1: Demo video for AYA Match app.Conceptualization, S.H.E.S., E.M.-H. and O.H.; Data curation, S.H.E.S. and M.J.P.R.; Formal analysis, S.H.E.S. and M.J.P.R.; Funding acquisition, S.H.E.S., E.M.-H. and O.H.; Investigation, S.H.E.S.; Methodology, S.H.E.S., E.M.-H. and O.H.; Project administration, S.H.E.S. and O.H.; Software, S.H.E.S.; Supervision, S.H.M.J. and O.H.; Visualization, S.H.E.S. and M.J.P.R.; Writing—original draft, S.H.E.S. and M.J.P.R.; Writing—review & editing, S.H.E.S., M.J.P.R., E.M.-H., B.V., P.P., S.H.M.J. and O.H. All authors have read and agreed to the published version of the manuscript.This research was funded by VIOZ-fonds (Stichting Vrienden Integrale Oncologische Zorg), Zorg Innovatie Fonds, en Janssen-Cilag B.V. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.The study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Institutional Review Board of The Netherlands Cancer Institute (IRBd19-089; 2019).Informed consent was obtained from all subjects involved in the study.The data presented in this study are available on request from the corresponding author.We thank IJsfontein GameWise B.V. for the process of building the application, Vivian Engelen (on behalf of Dutch Federation of Cancer Patient Organisation) for the elaboration of the needs assessment which led to the development of the app, Karin Hermans for the coordination of the developmental phase, and the AYAs and loved ones for openly sharing their experiences and their contribution to the development of the application.The authors declare no conflict of interest.Developmental stages of the co-creational process.During the paper prototype session AYA cancer patients and their loved ones were invited to interactively try out the concept.Flowchart of the different phases of the development of the application.Different screens of the final product: AYA Match. More is available in this demo video in the Supplementary Materials (Video S1).Quotes of AYAs and loved ones during the six-week test phase.1 All quotes in this table are translated from Dutch to English.Sociodemographic and clinical data of the Adolescent and Young Adult (AYA) cancer patients and loved ones using the AYA Match app.1 n/a = not applicable; 2 Participants were allowed to select multiple options, percentages do not add up to 100.Expectations of Adolescent and Young Adult (AYA) cancer patients and their loved ones before usage of the AYA Match App.1 Among the 121 participants, as described in Table 1, fourteen did not complete the concluding part of the questionnaire about their expectations of the AYA Match app; 2 Among the 37 participants, as described in Table 1, two did not complete the concluding part of the questionnaire about their expectations of the AYA Match app; 3 Participants were allowed to select multiple options, percentages do not add up to 100; 4 n/a = not applicable; 5 Additional reasons mentioned by respondents included: ‘I’ll accept all the help I can get’, ‘I’d like to get in touch with other patients’, ‘I’ve been asked to test the application by my patient organization’, ‘It’d be great to talk to someone who understands my situation’, ‘My doctor referred me’, ‘I have trouble asking for help and I hope this app will help’, ‘I’d like to communicate my needs to a group of people at once’, ‘I’d like to stay updated about my loved one with cancer’, ‘I mainly hope this app helps my AYA’.Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.