Dear Editor,
We read with great interest the article by Yim et al. [1], which explores the use of an immune signature score (ISS) and its refined 10-gene version (ISS10) to predict responses to Atezolizumab plus Bevacizumab (Atezo/Bev or A+T) therapy. The use of immune checkpoint inhibitors (ICIs) has revolutionized clinical care across cancer types through the potential for durable responses [2]. Atezo/Bev has dramatically altered the treatment landscape and is now the first-line treatment of advanced hepatocellular carcinoma (HCC) [3]. However, despite these advancements, the majority of HCC patients do not benefit, as the objective response rate (ORR) for single-agent ICI in HCC is around 15%, while the ORR for Atezo/Bev combination therapy remains at approximately 30% [4]. Therefore, there is an urgent need to develop predictive biomarkers for response to both ICI single-agent and combination therapies.
The response to ICI is influenced by the spatial and temporal variability of host, tumor, and tumor microenvironment (TME) [5]. While multiple predictive biomarkers for single-agent ICI response have been established, robust predictive biomarkers for Atezo/Bev are still lacking. Moreover, whether the immunotherapy response predictive signatures (IRPS) derived from other ICI agents and cancer types can be applied to Atezo/Bev in HCC remains largely unknown. To address this gap, we conducted a comprehensive analysis to identify Atezolizumab plus Bevacizumab response predictive signatures (ABRPS) based on transcriptomics.
First, we conducted an extensive literature review to gather ICI response indicators [6,7], resulting in the collection of 23 reported transcriptomic IPRS (Supplementary Methods and Supplementary Table 1). The size of these gene sets ranged from 2 to 186, with a median of 14 genes. Most of these signatures were derived from TME and exhibited some overlap (Supplementary Fig. 1). In our recent study, we developed a transcriptomic signature based on the fatty acid degradation (FAD) pathway, which has been shown to predict response to targeted therapy, transarterial chemoembolization (TACE), ICI monotherapy and combination therapies [8]. Notably, the FAD signature, which captures the metabolic features of tumor genomics, shares no common genes with other TME-based IRPS.
We then categorized HCC tumors within the IMbrave150 cohort into high and low signature subtypes based on the ssGSEA score of the corresponding signatures. Significant correlations were observed among the 23 IRPS, as shown by the correlation matrix calculated using ssGSEA scores (Supplementary Fig. 2A), revealing four distinct clusters (Supplementary Fig. 2B). Both the Cox proportional hazards model and Kaplan–Meier plots were employed to identify ABRPS (Supplementary Methods, Supplementary Fig. 3). Surprisingly, despite these 23 IRPS being derived from various cancer types and different ICI agents, 22 of them were identified as ABRPS when overall survival (OS) was used as the clinical outcome, and 15 were identified as ABRPS when progression-free survival (PFS) was used. Furthermore, 14 out of the 23 IRPS were identified as ABRPS even when applying a more stringent criterion that requires significance in both the Cox model and the log-rank test for both OS and PFS (Fig. 1A, 1B, Supplementary Fig. 3).
We further defined the hazard ratio (HR) score to evaluate and compare the predictive performance of the 23 IRPS in predicting response to Atezo/Bev, with a higher HR score indicating superior predictive accuracy. The ABRS (Atezolizumab plus Bevacizumab Response Signature), a 10-gene signature specifically developed from responders and non-responders in IMbrave150 trials [9], achieved the highest HR score, as anticipated. Notably, our FAD signature exhibited comparable efficacy (Fig. 1C). Interestingly, significantly different HR scores for OS and PFS were observed for signatures such as FAD, IMPRES, and C_ECM_ up, despite using the same dataset. This discrepancy may partly be due to these signatures identifying particularly aggressive phenotype of HCC (Supplementary Fig. 2). Additionally, we directly correlated these 23 signatures with the ORR to Atezo/Bev. Sixteen IRPS, including ABRS and FAD, showed a significant correlation with Atezo/Bev response (Supplementary Table 2). In contrast, only four IRPS, including the FAD signature, were associated with response to Sorafenib (Supplementary Table 3).
Identifying potential ICI predictive biomarkers is essential for minimizing immune-related adverse events and reducing treatment costs. We collected 23 transcriptomic biomarkers and showed that many of these IRPS, developed for other ICIs, are also predictive of response to the Atezo/Bev combination. We previously demonstrated that the HCC TME-based Inflamed signature, which captures the functional activity of immune cells [10], could predict response to anti-PD-1 therapy across various cancers [11]. Here, we have reaffirmed this concept using the Atezo/Bev dataset. Additionally, this study underscores the superior predictive performance of our HCC-cell-based FAD signature in predicting responses to Atezo/Bev.
There are several limitations of our study. Firstly, our analysis is based on a retrospective cohort. Secondly, despite our extensive literature review, it is possible that some relevant IRPS were overlooked. Thirdly, the precise underlying molecular mechanisms of ABRPS remain to be fully elucidated.
In summary, we conducted a comprehensive study to identify predictive biomarkers for Atezo/Bev, which not only complements but also extends the work of Yim. To the best of our knowledge, this is the first intergrated study to analyze and compare transcriptomic biomarkers for Atezo/Bev response. Further investigation in larger prospective cohorts is warranted to validated the ABRPS for Atezo/Bev in HCC.
PDF Links
PubReader
ePub Link
XML Download
Full text via DOI
Download Citation
Supplement1
Print
