Decreased Expression of T-Cell-Associated Immune Markers Predicts Poor Prognosis in Patients with Advanced Follicular Lymphoma Received Rituximab Plus Bendamustine

Background: Several clinical risk stratication models have been proposed to predict the clinical outcomes of follicular lymphoma (FL) cases, however, few reports are available to predict prognosis of FL cases receiving bendamustine-based regimens. We previously examined the utility of rituximab-bendamustine (RB) treatment for newly diagnosed advanced FL, who showed non-optimal responses to two cycles of R-CHOP therapy. Methods: In this study, we explored the biomarkers that could inuence outcomes for the RB-treated FL cases in the context of the prospective cohort by target capture and sanger sequencing, and gene-expression proling analyses using 50 diagnostic biopsies. Results: We rst examined the mutational status of 410 genes in tumor specimens derived from RB-treated cases. As reported before, CREBBP, KMT2D, MEF2B, BCL2, EZH2, CARD11, TNFRSF14, EP300, and APC were recurrently mutated, however, none of which was predictive for progression-free survival (PFS) in RB-treated cases. Similarly, the m7-FLIPI did not correlate with PFS or progression of disease within 24 months (POD24). A gene expression analysis using a panel of 770 genes associated with carcinogenesis and/or immune response showed that the expression of CD8+ T-cell markers (GZMM, FLT3LG, CD8A, CD8B, GZMK) and half of the genes regulating Th1 and Th2 responses were signicantly lower in the POD24 group than in the noPOD24 group. Finally, we selected 10 genes (TBX21, CXCR3, CCR4, CD8A, CD8B, GZMM, FLT3LG, CD3E, EOMES, GZMK), and dichotomized RB-treated cases into immune inltration high (inl high ) and inltration low (inl low ) clusters. The 3-years PFS rate was lower in the inl low cluster than in the inl high cluster (50.0% [95% CI: 27.1–69.2%] vs. EZH2, and T/NFSF14, excepting CARD11. These results raise the possibility that epigenetic regulation might be different between the two clusters. Of importance, the only signicant clinical parameter related to the inl low immune prole was lymphopenia, which we previously identied as a poor prognostic marker for RB-treated cases 21 . These results suggest that lymphopenia at diagnosis may reect reduced T cell-mediated immune reactions in the TME and that it may be able to substitute for the immune proling utilized in this study. exact test were shown in the graphs. (b) The proportion of cases with CARD11 mutations in the inllow and inlhigh clusters (n=40). p-values by Fisher’s exact test were shown in the graphs. Wt, wild type. Mut, mutations. (c)The proportion of cases with low absolute lymphocyte counts in the inllow and inlhigh clusters (n=48). p-values by Fisher’s exact test were shown in the graphs.


Background
Follicular lymphoma (FL) is the most common indolent subtype of B-cell non-Hodgkin lymphomas 1 . The majority of FL patients present with advanced stages 1 , and those with high tumor burden and/or diseaserelated symptoms require immunochemotherapy.
Given the heterogeneity of disease courses, several clinical risk strati cation models, including Follicular lymphoma International Prognostic Index (FLIPI-1 and FLIPI-2), PRIMA-PI, and FLEX were proposed to predict the clinical outcomes of FL patients such as progression-free survival (PFS) and overall survival (OS) [13][14][15][16] . However, all of these models have limited value to predict POD24. Meanwhile, accumulated ndings from genetic and gene-expression pro ling analyses showed that several molecular risk models mainly based on the intrinsic tumor properties, such as m7-FLIPI 17 , POD24-PI 18 and 23-gene-expression pro ling (GEP) score 19,20 , were shown to be a useful tool to identify FL patients at high-or low-risk of progression after treatment with immunochemotherapy. In addition, a gene expression signature based on the extrinsic tumor environment, particularly focusing on the tumor microenvironment (TME) of FL, was established to predict POD24 11 . However, these models are mainly based on the data from the patients who received R-CHOP or R-CVP, and few reports are available to predict the prognosis of FL patients receiving B-based regimens.
We previously conducted a multicenter prospective phase 2 (CONVERT) trial, and demonstrated the utility of RB for treatment-naive advanced FL showing non-optimal responses (nOR) to the initial two cycles of R-CHOP therapy 21 . In addition, we reported that peripheral lymphopenia (< 869/uL) before treatment was an independent poor prognostic factor for RB-treated FL patients in both our trial and validation cohorts 21 . Thus, the main object of this study is to identify molecular biomarker(s) that could predict outcomes of RB-treated FL patients by genetic and gene-expression pro ling analyses.

Study design and patients
The design and characteristics of the registered patients in the CONVERT trial were described previously 21  This study was designed and conducted according to the Declaration of Helsinki and was registered in UMIN (000013795) and jRCT (051180181). All patients provided written informed consent.
The protocol, informed consent forms, and any amendments were approved by the institutional review boards of each hospital and certi ed review board.
Target capture sequencing assay DNA was extracted from formalin-xed para n-embedded (FFPE) samples using a DNA storm kit (Cell Data Sciences, Inc). The targeted DNA library for panel sequencing that comprises approximately 1.65 Mb coding regions of all exons in 409 genes (Supplementary Table 1) was constructed using an Oncomine Tumor Mutation Load Assay (Thermo Fisher Scienti c) according to the manufacturer's protocol. In brief, 20 ng DNA was subjected to multiplex PCR ampli cation with an Ion AmpliSeq Library Kit 2.0. After multiplex PCR, Ion Xpress Barcode Adapters (Thermo Fisher Scienti c) were ligated to the PCR products, which were then puri ed with the use of AMPure XP beads (Beckman Coulter, Brea, CA). The puri ed libraries were pooled and then sequenced with an Ion Torrent S5 instrument and Ion 550 Chip Kit (Thermo Fisher Scienti c). DNA sequencing data were accessed through the Torrent Suite ver. 5.10 program (Thermo Fisher Scienti c). Reads were aligned against the hg19 human reference genome, and variants were analyzed with the use of Ion reporter ver. 5.10 (Thermo Fisher Scienti c). Raw variant calls were manually checked using the integrative genomics viewer (IGV, Broad Institute). In this analysis, we de ned the following criteria for the sequence quality (1) the uniformity of amplicon coverage of >60%, (2) the minimum quality criterion was 80% of target bases with ≥ 100x sequencing coverage. The mutations were ltered with the exclusion of (1) Fisher's exact P-value <0.01, (2) Phred QUAL score of <99, (3) allele frequency in tumor <0.025, and missense SNVs with an allele frequency of 0.45-0.55 in copy-neutral regions, unless they were listed in the COSMIC database (v70) or reported to be mutated in FL. Germline mutations were excluded with the NCBI dbSNP build 131 and UCSC genome browser database 23,24 .
Sanger sequencing analysis Exon 2 and 3 of MEF2B were analyzed using the PCR primers designed with Primer3 (Supplementary Table 2). The MEF2B coding exons 2 and 3 were ampli ed by PCR as described previously 25 . After denaturation step at 94°C for 5 min, PCR consisting of denaturation (94°C for 30 sec) annealing (56°C for 30 sec), and extension steps (72°C for 30 sec) were repeated until 35 cycles, followed by nal extension step at 72°C for 7 min on a GeneAmp PCR system 9700 (Thermo Fisher Scienti c). After visual con rmation of ampli cation, the PCR product was submitted for sanger sequence analysis as previously described 25 .
The m7-FLIPI score calculation The m7-FLIPI score was calculated by using the calculator presented at the German Low-Grade Lymphoma Study Group o cial internet site (www.glsg.de/m7-ipi/). The same cut-off level as used in the original publication was applied to determine high-and low-risk categories (m7-FLIPI score ≥0.8 and <0.8, respectively) 17 .
Gene-expression pro ling analysis RNA was extracted from FFPE samples using an RNA storm kit (Cell Data Sciences, Inc). Details of the nCounter assay (NanoString Technologies, Seattle, WA, USA) have been reported previously 24,26 . In this study, the customized PanCancer immune-pro ling panel (NanoString Technologies), which consists of 770 genes related to cancer or immune cells and additional 30 genes (Supplementary Table 3), was used for nCounter-based gene-expression measurements. The data were analyzed by using nSolver 4.0 software (NanoString) and JMP version 13.0 software (SAS Institute).

Statistical analysis
The cutoff date for this analysis was December 31, 2019. The survival analysis was performed by the Kaplan-Meier method, which was evaluated by a log-rank test. Fisher's exact test was used for between-categorical data comparisons. For the comparison of two continuous variables data were tested by Wilcoxon rank sum test. Statistical analyses were performed with R 4.1.0 software (The R Foundation for Statistical Computing), and p value <0.05 was considered statistically signi cant.

Mutational landscape of the whole population
To investigate the mutational landscape in this cohort, we performed the target capture sequence using tumor samples from the above described 50 cases. Eight cases were excluded from this analysis because of their poor DNA qualities. Thus, 42 cases (OR n=8; nOR n=34) were analyzed on the targeted panel for 409 genes, in which most of the recurrently mutated genes in FL were included excepting MEF2B. MEF2B encodes a transcriptional activator and is mutated in 10∼20% of FL and its mutational status is applied in m7-FLIPI 27 . Thus, we analyzed its mutation status by sanger sequencing independently. In this analysis, we focused on the exon 2 and exon3 of MEF2B, because most of MEF2B mutations in FL were scattered around these regions 28 . As a result, MEF2B mutations were identi ed in 13/42 cases (31.0%), all consisting of L67R in the exon 3, which mutated site was reported to play an important role in the interaction with EP300 proteins 28 . The double mutations with L67R and G2R in exon 2 were identi ed in one case (Fig. 1a).
Association of mutation pro le with clinical outcomes in the nOR-group Among recurrently mutated genes, their mutation frequencies (in ≥10%) in the OR-and nORgroups were shown in Table 2. Next, we analyzed whether mutation status in uenced the outcomes in the nOR-group (n=34). As shown in Fig. 2a, the mutation status of the speci c gene did not in uence PFS signi cantly in univariate analyses. We also evaluated the in uence of gene mutations on POD24. Among 7 cases with POD24 and 27 cases without POD24, 28.6% (2/7) of POD24 cases harbored PTEN or CIC mutations as compared with 0% (0/27) in cases without POD24, with a statistical difference (p=0.0374) (Fig. 2b). Interestingly, no POD24 event occurred in RB-treated cases with CARD11 mutations in the nOR-group. In addition, the cases with CARD11 mutations tended to have better 3-years PFS rates than unmutated cases but with no statistically signi cant difference (100% vs. 59.3%, p=0.174) (Fig. 2c).

Association of the 23-gene expressions with clinical outcomes
We next analyzed gene expression levels by nCounter system using the FFPE samples from 48 cases. Sample quality was assessed by mRNA levels of 40 housekeeping genes in each sample (Supplementary Table 3). It was previously reported that the GEP score based on the expression of 23 genes was associated with clinical outcomes of FL cases 19 , which was con rmed by the following study 20 . Furthermore, the 23 genes could identify two main clusters characterized by high-or lowexpression associated with favorable and poor outcomes 20 . So, we subjected our 48 cases to hierarchical clustering using these 23 genes. Consistent with the previous reports 19,20 , we could identify two main clusters (cluster 1 and cluster 2) (Fig. 4a) (Fig. 4c). We further compared the expression of 23 genes between 8 cases with POD24 and 31 cases without POD24. Although these genes were reported to portend good or poor risk, the expression levels of these genes were almost equivalent in the two groups ( Supplementary Fig. 3).

Association of the T-cell-associated gene expressions with clinical outcomes
In this trial, we previously reported that a low absolute lymphocyte count (ALC) was an independent poor prognostic factor for RB-treated patients 21 . Because lymphopenia was considered to re ect an immune-suppressive state, we applied the customized pan-cancer immune-pro ling panel, which consists of 770 genes related to cancer development and/or immune reactions, to the RB-treated cases in the nOR-group (n=39). As a result, a total of 33 genes, including 2 upregulated (CD79a and POU2F2) and 31 downregulated genes (each with log2 fold change > |0.5|, and corrected [-log10P] >1.5) were differentially expressed in the POD24 group (n=8) compared with in the noPOD24 group (n=31). Of interest, the top genes downregulated in the POD24 group were markedly enriched with the T-cellassociated genes (Fig. 5a).
Thus, we further compared the expression levels of molecules speci c for various types of T cells (Th1, Th2, and Th17, T follicular helper [Tfh], regulatory T [Treg], NK, and CD8 + T cells) and of immune checkpoint markers between the two groups. Interestingly, the expression of CD8 + T cell markers (GZMM, FLT3LG, CD8A, CD8B, GZMK) except for PRF1 was signi cantly lower in the POD24 group than in the noPOD24 group (Fig. 5b). Furthermore, about half of the genes related to Th1 and Th2 responses in T cells were downregulated in the POD24 group compared to those in the noPOD24 group. In contrast, Treg, NK, and immune checkpoint markers were almost equivalently expressed in the two groups ( Supplementary Fig. 4).

Association of the T-cell-associated gene expressions with their mutation pro le
We nally analyzed whether gene mutations in uence the expression of T-cell-associated genes in all cases. However, the frequencies of mutations were roughly the same between in l low and in l high clusters in most genes (Fig. 6a). For example, the mutation frequencies in CREBBP 31 , EZH2 32 , and T/NFSF14 33 , which are supposed to regulate the tumor microenvironment, did not differ signi cantly between the two clusters. In contrast, a signi cant difference was noticed for the CARD11 mutations. In the in l high cluster, 31.2% (6/19) of the cases harbored CARD11 mutations compared to 4.76% (1/21) in the in l low cluster (p=0.0395) (Fig. 6b). Of note, the proportion of cases with low ALCs was higher in the in l low cluster than in the in l high cluster (38.5% vs. 9.09%, OR: 6.25 [95%CI, 1.20-32.7], p=0.0235) (Fig.  6c).

Discussion
In this study, we attempted to identify biomarker(s) that can predict clinical outcomes of RB-treated FL cases. At rst, we examined gene mutational pro les in our cases. As a result, we found that types of mutated genes and their frequencies were roughly the same as previously reported 1,29,30 , and we found that none of those mutations in uenced PFS in univariate analyses. Among them, mutations of CARD11, which functions as a positive regulator of the NF-kB pathway in normal B and T lymphocytes, were detected in 16.7% of the cases (7/42). CARD11 mutation is more frequently observed in transformed FL than in untransformed FL 29,34 and counted as a poor prognostic factor in m7-FLIPI 17 . In accord with these results, all cases harboring CARD11 mutations didn't show optimal responses to the initial 2 cycles of R-CHOP and were classi ed into the nOR-group. However, all these cases achieved and maintained CR at least for 3 years by the early conversion to RB treatment. These results raised a possibility that the negative impact of CARD11 mutations observed under R-CHOP treatment might be canceled by RB treatment. Meanwhile, 28.6% (2/7) of RB-treated cases with POD24 harbored PTEN or CIC mutations as compared with 0% (0/27) in cases without POD24 with a statistical difference (p = 0.0374). However, further studies are necessary to elucidate the clinical signi cance of PTEN or CIC mutations in RB-treated FL cases.
We also assessed whether m7-FLIPI is useful to predict prognosis of our RB-treated FL cases. Although the 3-years PFS tended to be higher in the m7-FLIPI low group than in the m7-FLIPI high group (70.8% vs. 60.0%), this difference was not signi cant (p = 0.264). Also, whereas m7-FLIPI was reported to be able to predict POD24 in R-CHOP-treated FL patients, POD24 was observed regardless of m7-FLIPI risk groups in our BR-treated cases (high 30% vs. low 16.7%, p = 0.394). It must be noted that although the results reported in this study may lack statistical power due to the small number of cases. In accord with our results, however, m7-FLIPI had no prognostic value in RB-or GB (G + B)-treated FL cases, as reported in a randomized phase III (GALLIUM) trial 3,35 . Together, these results suggest that m7-FLIPI may not be an accurate prognostic factor in FL cases treated with B-based regimens.
The GEP score based on the expression of 23 genes was initially established to predict clinical outcomes in FL patients 19 , and based on the 23-gene expression panel, it was possible to identify two main clusters associated with favorable and poor outcome in the subsequent study 20 . However, these studies included only a few cases treated with RB. Subsequently, this model was shown to have no prognostic value in RBor GB-treated cohort in the GALLIUM study 36 . In accord with this result, we found that the 3-years PFS rates didn't differ between the cases in cluster 1 and cluster 2 (72.4% vs. 50.0%, p = 0.263). However, cluster 2 cases tended to have an inferior 3-years PFS rate compared with cluster 1 cases. Considering the small number of cases in this study, further studies are needed to draw a de nite conclusion as to the e cacy of this model in predicting the prognosis of B -treated FL cases.

Several lines of evidence indicate that tumor cells depend on the interactions with non-malignant cells
that constitute TME for their growth and survival 1,37 . Particularly, because of the indolent feature of FL cells, the prognosis of FL is substantially affected by TME, which consists of various types of T lymphocytes (cytotoxic T cells [CTLs], Tregs, and so on), B lymphocytes, and tumor-associated macrophages (TAMs). These cells act as positive or negative regulators of immune response in FL, thereby in uencing the clinical outcomes in this disease. Based on this concept, Tobin JWD, et al. analyzed the immune in ltration pro le in FL cases with or without POD24, and observed that a low expression of immune effectors (TNFα, CD4), checkpoints [programmed death-ligand 2 (PD-L2)], or macrophage markers (CD68), were associated with POD24. Among them, PD-L2 was the most sensitive marker in identifying patients with poor prognosis in R-CHOP-treated patients 11 .
To elucidate the roles of the immune microenvironment in RB-treated FL cases, we have applied the customized pan-cancer immune-pro ling panel consisting of 770 genes to the RB-treated cases in the nOR-group. As a result, we found that CD8 + T cell markers such as CD8A, CD8B, FLT3LG, GZMM, and GZMK, were downregulated in the POD24 group, while markers for Treg and NK/T cells and immune checkpoint molecules were almost equivalently expressed in cases with or without POD24. Among the genes downregulated in the POD24 group, GZMM and GZMK, which are commonly expressed in CTLs and NK cells, have pro-apoptotic activities on tumor cells 38 . A low expression of GZMM and GZMK in tumor tissues was associated with an unfavorable prognosis in cutaneous melanoma 39 , and GZMK was also found to be downregulated in the process of FL transformation 40 . Together, these results suggest that anti-FL immune responses may be weakened in RB-treated FL cases undergoing POD24. However, future studies using high throughput technologies, such as single cell transcriptome sequencing, multicolor immunohistochemistry, and imaging mass cytometry are needed to determine which T cell subpopulations indeed express each cytotoxic molecule.
We further classi ed RB-treated cases into in l high and in l low clusters using manually selected 10 genes (TBX21, CXCR3, CCR4, CD8A, CD8B, GZMM, FLT3LG, CD3E, EOMES, GZMK). The PFS rate at 3 years was signi cantly lower in the in l low cluster than in the in l high cluster (p = 0.0237). In addition, POD24 was observed in 8/20 (40%) cases in the in l low cluster, while none of in l high cases underwent POD24 (p = 0.0084). These results suggest that our classi cation would be useful to predict clinical outcomes including POD24 in RB-treated FL cases. Also, given the results presented here, the combination of RB with a novel therapy, which can restore the number and/or function of T cells such as lenalidomide 41,42 , would be a promising strategy to improve prognosis of the cases with in l low immune pro le. However, it should be noted that our RB-treated cases constitute a unique population, namely those who received RB after two cycles of R-CHOP. Therefore, future studies are required to determine whether our results are applicable to de novo FL cases treated with a B-based regimen as the 1st -line treatment using larger prospective cohorts.
To clarify molecular mechanisms underlying the difference in in l low and in l high immune pro les, we compared the frequency of mutations in both clusters. However, the frequencies of mutations were roughly the same between in l low and in l high clusters in most genes, especially that regulate TME such as CREBBP, EZH2, and T/NFSF14, excepting CARD11. These results raise the possibility that epigenetic regulation might be different between the two clusters. Of importance, the only signi cant clinical parameter related to the in l low immune pro le was lymphopenia, which we previously identi ed as a poor prognostic marker for RB-treated cases 21 . These results suggest that lymphopenia at diagnosis may re ect reduced T cell-mediated immune reactions in the TME and that it may be able to substitute for the immune pro ling utilized in this study.

Conclusions
We here found that the expression of CD8 + T cell markers was signi cantly lower in RB-treated cases with POD24 than those without POD24. Furthermore, FL cases were classi ed into in l low and in l high clusters based on the expression of T cell-associated genes, which was useful to predict the prognosis of RBtreated FL cases. 39. Wu X, Wang X, Zhao Y, Li K, Yu B, Zhang J. Granzyme family acts as a predict biomarker in cutaneous melanoma and indicates more bene t from anti-PD-1 immunotherapy. Int J Med Sci. 2021;18 (7)