Developments and emerging trends in PD-L1 research in gastrointestinal cancers (2000-2018): a bibliometric perspective

Background: The programmed death-ligand 1 (PD-L1) pathway inhibits T-cell receptor-mediated production of IL-2 and T-cell proliferation and plays an important role in the immunosuppression of various types of cancers. An increasing number of studies have focused on the potential utilization of anti-PD-L1 therapy in gastrointestinal cancers. In this study, we aimed to analyze developments and emerging trends in studies of PD-L1 in gastrointestinal cancers from a bibliometric perspective. Methods: Manuscripts were retrieved from th Web of Science Core Collection (WOSCC) Database. CiteSpace, a bibliometric software, was used to identify landmark studies, key concepts, and various subtopics in this research area. Results: A total of 1325 manuscripts examining PD-L1 in gastrointestinal cancers were included. Manuscripts published in 2017 and 2018 accounted for almost half of the publications (44.2%, 586/1325). Combined with 31,960 references, the manuscripts on this topic constituted a complex co-citation network, and landmark papers were identified by indexes including citation in the network, betweenness centrality, and burstiness . Key concepts such as “Regulatory T cell,” “TIL,” and “Her2” were identified in the co-citation network for author keywords. Furthermore, several subtopics were identified during the process of “clustering” in colorectal, gastroesophageal, and hepatopancreatobiliary cancers, such as “predictive biomarkers”, “advanced cancers”, and “clinical efficacy”. Conclusions: Research on PD-L1 i n gastrointestinal cancers is a rapidly progressing area. More scientific findings are expected in the near future. Analysis and summarization from a bibliometric perspective not onlyidentify landmark manuscripts and hot-spot concepts but also indicate possible directions for future studies.


3
Immunotherapy plays an important role in clinical cancer therapy, and many checkpoints have been discovered for tumor suppression. Immune checkpoint therapy, mainly antiprogrammed death-1 (PD-1) and PD ligand 1 (PD-L1) therapy, can enhance antitumor immune responses by blocking the inhibitory signals of the immune system. PD-L1, also known as CD274, was first discovered by Dong et al [1] in 1999 as an immune regulatory molecule called B7-H1. Later, B7-H1 was renamed PD-L1 because it was identified as the ligand of PD-1. Moreover, blockade of B7-H1 reduced the growth of tumors in the presence of immune cells. Importantly, Dong et al [2] published a landmark paper reporting that tumors with PD-L1 expression had increased T-cell apoptosis. Engagement of PD-L1 with its receptor PD-1, which is expressed on T cells, inhibits T-cell receptor (TCR)-mediated activation of IL-2 production and T-cell proliferation. The interaction between PD-L1 and PD-1 also contributes to ligand-induced TCR downmodulation during antigen presentation to naive T cells [3]. This significant finding promoted PD-L1 as a potential target in cancer immunotherapy. Currently, several PD-L1 antagonists (avelumab, durvalumab, and atezolizumab) are approved by the US Food and Drug Administration for various indications, especially in treating non-small cell lung cancer (NSCLC) and melanoma [4][5][6][7].
Several high-quality clinical studies have examined antagonists of PD-1 in various gastrointestinal cancers. El-Khoueiry et al [8] reported that nivolumab had a manageable profile and durable objective responses in advanced hepatocellular carcinoma. Shitara et al [9] assessed combination pembrolizumab and paclitaxel to treat gastric or gastroesophageal junction cancer. Although no significant improvement of overall survival was observed for this new strategy, pembrolizumab had a better safety profile than Several reviews have already been published on the frontier PD-L1 research in clinical applications. However, analysis with a bibliometric method, which relies on an artificial intelligence-based algorithm to landmark manuscripts, hotspots, and emerging trends in a research area, has not been done. In this study, we used CiteSpace [11], a bibliometric software, to summarize current research about PD-L1 in gastrointestinal cancers and explore possible directions for future studies.

Overview of landmark manuscripts
The 1325 manuscripts about PD-L1 in gastrointestinal cancers also cited each other or some other literature in their references. Combined with their 31,960 references, a simplified co-citation network ( Figure 3) of the manuscripts was constructed on the condition of the g-index as 5. In Figure 3, the 5 manuscripts according to the value of total citations were marked. They were summarized in Table 1 Table 2.

Key concepts in the research area
To identify key concepts about PD-L1 in gastrointestinal malignancies, a foam tree was constructed using the Carrot2 analytic software. The top 5 keywords with the highest frequency were circled with red line, as "gastric cancer," "therapies for the treatment," "colorectal cancer," "PD-1 for the treatment," and "CD8 T cells and tumor" (Figure 4).

Topics in various gastrointestinal cancer types
To understand different subtopics in various gastrointestinal cancer types, a unique process termed "clustering" was utilized. Simplified co-citation networks (restriction on the condition of the g-index as 5) were constructed for colorectal, gastroesophageal, and hepatopancreatobiliary cancers, and studies in these three cancers were divided into 32, 23, and 25 clusters, respectively ( Figure 6). The three largest clusters in each disease were summarized from the titles of manuscripts in the cluster (Table 4), which reflected the contents of each cluster. For example, in colorectal cancers, the terms summarizing the largest three clusters were "predictive biomarkers," "advanced cancer," and "prognostic significance."

Discussion
The clinical utilization of PD-L1 antagonists initially focused on NSCLC and melanoma.
Herbst et al [49] performed a randomized controlled trial comparing pembrolizumab and docetaxel for treating NSCLC, which suggested that pembrolizumab prolongs overall survival in patients with previously treated PD-L1-positive NSCLC. Buchbinder et al [50] found that a high dose of IL-2 enhanced the effect of PD-L1 inhibitor in melanoma patients. From 2000 to 2018, an increasing number of studies were conducted to explore the mechanism and clinical utilization of PD-L1 antagonists in gastrointestinal cancers, which possibly provided new strategies for immunotherapy in the future.
Although several reviews have already discussed the current state of PD-L1 in gastrointestinal cancers, a systemic analysis of all literatures and correlated manuscripts in this area was lacking [51,52]. According to our results, over 1300 manuscripts have been published on this topic, with over 30,000 combined references. The citations between these papers constitute an extremely complex co-citation network. Therefore, traditional reviews, which are mainly based on expert opinions, are limited in their ability to summarize such difficult correlations. Here, we used the bibliometric software CiteSpace, which is user-friendly and easily mastered by researchers without a scientometric background, to summarize the current state of this field. In this study, combined with data analyses from the WOSCC Database, several important findings were made. Third, by analyzing author keywords and "clustering," some important phrases were identified. In the co-citation network of author keywords, "TIL," "Her2," and "Radiation" were identified to have high burstiness. TILs, or tumor infiltrating lymphocytes, are considered predictive markers when combined with PD-L1 expression. In the study by Cariani et al [55], PD-L1 combined with TILs could better predict the prognosis of hepatocellular carcinoma. Her2 is a frequently measured molecular marker in various types of cancers. In the study by Li et al [39] on gastric cancer, the expressions of Her2 and PD-L1 were correlated, indicating the possibility of a combined therapeutic strategy.
Radiation together with chemotherapy was of clinical efficacy in rectal cancer. In Hecht's study [40] on rectal cancer, PD-L1 was upregulated by chemoradiotherapy, suggesting the possibility of using a PD-L1 antagonist with traditional radiation or chemotherapy. During the process of clustering, some phrases were also identified. For instance, the terms "advanced cancer", "metastatic gastric cancer" were identified. Currently, most of clinical studies focusing on immune checkpoints were done in patients with advanced cancers [8,9,49]. Whether these medications would be useful for patients at eariler stage or after curative resection need to be validated in future. There were mainly two limitations of this study. One was that the co-citation analyses were only done in papers derived from the WOSCC. Results from other databases such as PubMed, Scopus were not seperately analyzed. The other was that among all the bibilometric softwares, mainly CiteSpace was utilized. The software of VOSviewer, CitNetExplorer and HistCite could provide similar partial function of CiteSpace. However, the visuliaztion of science map based on CiteSpace was easiler to be understood by researcher without background of scientometrics.
The CiteSpace V 5.3.R4 (64 bits), a bibliometric application invented by Professor Chaomei Chen [11], was utilized for most of the intellectual analyses in this study. Several functions were utilized. First, to identify landmark papers, a co-citation network was constructed, in which each node represents a manuscript (including the reference paper) and each link represents the relation of citing. The network could be constructed in a completed manner to include all the manuscripts and references, or in a simplified manner that neglects studies with a minimum chance of citations (restriction with g-index [56]; the scale factor k was set as 5). In co-citation networks, the importance of each node could be measured by its total citation amount in the network or the value of betweenness centrality [57] and burstiness [58]. In this study, a landmark manuscript was defined as a paper ranked top 5 according to these three indexes.
Second, to identify key concepts evolving through time, author keywords in all manuscripts were utilized to construct a co-citation network and simplified by the condition that only phrases with a top 50 citation amount were used per year. Then, the burstiness of each keyword was measured. High burstiness indicated that the citation counts sharply increased over time. The keywords with high burstiness and scientific significance were analyzed. Additionally, we used the software Carrot2 [59] to identify important keywords directly from the title of each manuscript. The results of the Carrot2 analysis were visualized as a "foam tree." Third to understand different topics in various gastrointestinal cancers, a unique process called "clustering" was performed in the simplified co-citation networks (restriction of gindex of 5) of colorectal, gastroesophageal, and hepatopancreatobiliary cancers. The quality of clustering was measured by two indexes, modularity and silhouette score [60].
Finally, the label of each cluster was summarized with the method of log-likelihood, which represents the subtopics of each disease.

Competing interests
The authors declare that they have no competing interests.

Authors' contributions
TNY and XJC conceived the study and performed critical revision of manuscript. TNY, GYJ, XYL, HL and QZ designed the study, performed statistical analyses and drafted the manuscript. GYJ and TNY wrote the manuscript. TNY and XJC performed the article retrieval, data interpretation and provided supervision. All authors read and approved the final manuscript

Availability of data and materials
The datasets during and/or analyzed during the current study are available from the corresponding author on reasonable request.    Annual publications about pd-l1 in gastrointestinal cancers Figure 1 Annual publications about pd-l1 in gastrointestinal cancers Key concepts about pd-l1 in gastrointestinal cancers (circled in red line) Clustering in colorectal, gastroesophageal, and hepatopancreatobiliary cancers Figure 6 Clustering in colorectal, gastroesophageal, and hepatopancreatobiliary cancers