B cells sustain inflammation and improve survival in human melanoma

Tumor associated inflammation is one of important predictors of response to immune checkpoint blockade. Understanding molecular processes that regulate tumor inflammation is key to improve the efficacy of checkpoint blockade. Established mechanisms that underlie therapy response and resistance have centered on anti-tumor T cell responses. We show that tumor-associated B cells are vital to T cell functions. They promote recruitment of CD8+ T cells through plasmablast-like cells with expression of pro- and anti-inflammatory factors. Plasmablast-like cells are associated with improved survival of patients with metastatic melanoma and their response to checkpoint blockade. Plasmablast-like B cells express chemokines for T cell-attraction. Depletion of tumor-associated B cells by anti-CD20 immunotherapy of metastatic melanoma patients causes a remarkable decrease in tumor CD8+ T cells. These findings indicate that tumor-associated B cells orchestrate and sustain tumor inflammation, recruit CD8+ T effector cells and are key to therapeutic response and patients’ overall survival.


Introduction
Cancers such as melanoma, lung, and kidney cancer often present with an inflamed but immunosuppressive tumor microenvironment (TME). Immune checkpoint blocking (ICB) antibodies have significantly improved cancer therapy by overcoming inhibition of T cell effector functions. Yet, a considerable number of patients does not benefit from ICB therapy 1 . It is therefore key to understand the mechanisms that regulate inflammation within the TME to develop novel therapies and improve patient survival.
B cells promote both acute immune-associated inflammation for protection against foreign pathogens as well as chronic inflammation in autoimmune diseases and persistent infection. Mouse cancer models show that tumor-associated B cells (TAB) promote tumor inflammation 2,3 but may also inhibit anti-tumor T cell-dependent therapy responses [4][5][6][7] . The immuno-inhibitory function of TAB in these models resembles that of regulatory B cells (Breg), which are an established source of inhibitory cytokines such as IL-10 and TGF-b (reviewed in 8 ). In human cancer, Breg were described by either phenotyping, direct detection of immunoinhibitory cytokines or surface molecules, and/or immunosuppressive function 4,[9][10][11][12][13] .
Often Breg frequencies increase with tumor progression and are enriched in tumors compared to peripheral blood or adjacent normal tissue. Increased IL-10 + B cell numbers can also be accompanied by increased numbers of CD4 + CD25 +/high CD127 low/and Foxp3 + Tregs in tumor tissues 10,12,14,15 which were independently associated with tumor progression or reduced patient survival.
In human melanoma, up to 33% of the immune cells can be TAB 16,17 and phenotypic analysis has revealed CD20+ TAB (reviewed in 18 ) and CD138 + or IgA + CD138 + plasma cells 17,19 .
Conclusions about their impact on disease progression and outcome are inconsistent. So far, no data exist for TAB functions in murine models of melanoma highlighting the need for studies in melanoma patients and tumor samples. We recently showed that TAB-derived IGF-1 is a source of acquired drug resistance of human melanoma to mitogen-activated protein kinase (MAPK) inhibitors 16 . Clinical data from our pilot trial and an independent case series indicate objective tumor responses and clinical benefit through B cell-depletion by anti-CD20 antibodies in end-stage therapy-resistant metastatic melanoma patients 16,20 .
Human melanoma cells foster the generation of TAB with regulatory activity [21][22][23] . They provide antigens for prolonged B cell receptor stimulation and release inflammation-modulating cytokines such as IL-1b, IL-6 24 and IL-35 25  TAB are primarily located at the invasive tumor-stroma margin 16,17,26 . This suggests a preferentially contact independent communication between melanoma cells and TAB.

Human melanoma cells directly induce NFKB activation in TAB through soluble factors
The known release of pro-and anti-inflammatory cytokines from melanoma cells and our multiplex immunostaining results suggest that melanoma cells communicate through soluble factors with TAB. We therefore exposed immortalized peripheral blood-and tumor-derived B cells derived from 4 patients with metastatic melanoma to melanoma-conditioned medium collected from autologous early passage melanoma cells. In a screen for expression of the key immunoregulatory and pro-inflammatory cytokines/molecules PD-L1 and IL-6, all melanoma conditioned media induced a similar up-regulation in matched peripheral blood-and tumor-derived B cells (Supplementary Figure 4A). We therefore performed subsequent experiments with the melanoma conditioned medium from patient 2 as it induced the most prominent changes.
572 genes/proteins were identified as significantly regulated by melanoma conditioned medium in both, the RNA-seq and proteomics results. Additionally, 974 proteins were found only significantly regulated in the proteomics data and 2,520 genes found only significantly regulated in the RNA-seq data (Benjamini-Hochberg (BH) corrected p-value < 0.05, Supplementary Table   1). There was no marked difference between peripheral and tumor-derived B cells (Figure 2A).
The estimated fold changes of genes identified as significantly regulated in both approaches showed a high linear correlation (Spearman cor = 0.79, p < 0.01, Supplementary Figure 4B).
The subsequent pathway analysis (see Methods) showed upregulation of pathways associated with inflammation, immunity, B cell receptor signaling and intracellular signal transduction, and downregulation of pathways associated with cell cycle, cell division, DNA replication, DNA repair, translation, and transcription (Supplementary Table 1).
One of the most significantly upregulated pathways was "tumor necrosis factor (TNF) signalling via NFKB. NFKB is a key transcription factor for B cell activation in inflammation and immune response 27 . This pathway includes, among others, CD69, CD80, CD30 (TNFRSF8), and CD137 (4-1BB/TNFRSF9) which were significantly upregulated in our proteomics and transcriptomics data (with CD30 (TNFRSF8) among the top-20 upregulated genes) and are known to be upregulated in B cells upon activation. Even though TNF itself showed no significant differences, TNF alpha induced protein 2 (TNFAIP2) was significantly upregulated as indirect evidence for TNF signalling. In addition, we observed the up-regulation of CD40 signalling genes which also acts via NFKB. Additionally, we found increased phosphorylation of PRKCB which plays a key role in B cell activation by regulating BCR-induced NFKB activation (with no significant differences in transcriptomics or global proteomics analysis). Similarly, we observed an increased phosphorylation of NKAP which is involved in TNF-and IL-1 induced NFKB activation. Therefore, melanoma-derived soluble factors directly induce a signalling pattern in B cells associated with activation in inflammation and immune response.

Melanoma conditioned medium induces a plasmablast-like dominated TAB population
The downregulation of cell cycle associated pathways coincided with a significant, unexpected down-regulation of CD20 (MS4A1, BH adjusted p < 0.01 RNA-seq) and CD19 (BH adjusted p < 0.01 proteomics). This indicates a phenotypic change of B cells. To identify this phenotype, we first used the Reactome functional interaction network 28   We used scRNA-seq data from two recent large studies on immune-checkpoint response in melanoma 29,30 to validate these functional signatures. Since CD19 and CD20 were downregulated in the induction experiments, we compared CD27+CD38+ TIPB with CD19+CD20+ TAB ( Figure 3A, 3B). This resulted in 623 CD19+CD20+ TAB and 814 TIPB (Jerby-Arnon et al. 30 ) and 1,063 CD19+CD20+ TAB and 2,941 TIPB (Sade-Feldman et al. 29 ).
The data scarcity prevented the definition of more fine-grained phenotypes. Only 5% of cells expressing CD27 and CD38 were also expressing CD138 (SDC1) in both datasets. 40  We defined a cell to express a signature if it expressed at least all of the signature's genes minus one with a minimum number of required genes of two and a maximum of four. This analysis showed that the activation, co-stimulation and immune-checkpoint signatures are expressed in TIPB and to a lesser extent in CD19+CD20+ TAB (Fisher-exact test, BH corrected p-value < 0.01 both datasets, activation Jerby-Arnon BH adjusted p = 0.03, Figure 3C). Both, CD19+CD20+ TAB and TIPB expressed substantial levels of immunosuppressive cytokines such as TGFB1 and IL-10 ( Figure 3D). Expression of the latter in TIPB could be additionally validated through multiplex immunostaining ( Figure 3E). Remarkably, TIPB also expressed numerous chemokines including T cell and macrophage chemoattractants XCL1, CCL22 and CSF1 ( Figure 3D). Together, these results indicate that TIPB are able to regulate inflammation and shape the cellular composition of the human melanoma TME. additionally showed a high linear correlation with our TIPB signature (Spearman correlation >= 0.85, BH adjusted p < 0.01, Supplementary Figure 5D). Finally, high expression (above median) of our TIPB signature was correlated with longer overall survival in the TCGA cohort ( Figure   4A). These results show, that the TIPB and the predicted functional signatures, are highly associated with T cell abundance and inflammation, and, importantly, are linked with patient outcome.
Anti-PD1 therapy frequently leads to an increase in B cell numbers which should enhance our functional signatures. We used the transcriptomics data by Riaz  versus lower 25%) predicted overall survival (Likelihood ratio test p = 0.06, Figure 4C).
Together, these data validate the clinical importance of the identified TIPB.

Loss of TAB reduces inflammation in the tumor microenvironment
Finally, we evaluated the loss of TAB in a cohort of patients with metastatic melanoma treated with anti-CD20 antibodies 16 Principal component analysis of whole tissue RNA-seq showed no systematic difference between the two patient groups ( Figure 5A).
18 The "tide model" describes expression of co-stimulatory and inhibitory signals as a dynamic system to fine-tune immune responses 49 . Additionally, the "immune set point model" highlights the often subtle contribution of intrinsic and extrinsic factors that modulate tumor inflammation and tip the balance between immunity and tolerance 38

Competing interests
The authors declare no competing interests.

Patient-derived material
Of 10 patients with metastatic melanoma who were treated with the anti-CD20 antibody ofatumumab in a therapeutic setting 16 Table 3.

Induction experiments with Melanoma-conditioned medium (MCM)
Immediately immortalized TAB has revealed significant differences predominantly in pathways related to cell cycle proliferation, apoptosis, and interferon response 52

Cell lysis and protein digestion
All reagents were of analytical grade and obtained from SIGMA-Aldrich, unless specified otherwise. Cells were lysed in freshly prepared lysis buffer containing 100 mM Tris/HCL pH 7.6, 2 % sodium dodecyl sulfate (SDS), 1 mM sodium vanadate, 1 mM NaF, protease inhibitor (cOmplete Tm EDTA-free) and phosphatase inhibitor (PhosSTOP Tm ) cocktail tablets (both Roche).
Cell extraction and DNA sharing was assisted by sonication and cell debris pelleted by centrifugation at 20.000 x g for 15 min at 20°C. The supernatant was collected and the total protein concentration determined using the BCA protein assay kit (Pierce Biotechnology).
Filter-aided sample preparation (FASP) was performed using Amicon Ultra Centrifugal 30 kDa molecular weight cutoff filters (Millipore) essentially according to the procedure described by 55 .
In between, beads were spun down and the supernatant was discarded.

Transcriptomics analysis
Library preparation and sequencing The amount of total RNA was quantified using the Qubit Fluorometric Quantitation system (Life Technologies) and the RNA integrity number (RIN) was determined using the Experion Automated Electrophoresis System (Bio-Rad).
For the co-culture experiments the isolated RNA (1µg) was processed using the SENSE

Data analysis and statistical information
All statistical tests were performed using R version 3.5.1 58  Signature expression levels were estimated using the ssGSEA approach and cell type abundances estimated using xCell (see above).
In a second step, multiplex immunostains were established essentially as described 77,78 .
Random integration of sequential Abs within a multiplex panel may lead to imbalanced signals, incomplete staining through interference with previously applied tyramide signal amplification (TSA), disruption of epitopes, and removal of TSA fluorophores because of repetitive antigen-retrievals at high temperature 77,78 . Therefore, each Ab was tested individually for its optimal position in the sequence of multiplex staining to minimize interference with previous Ab-TSA complexes or by alteration of epitopes.
For multiplex immunostains, 4 μm sections were deparaffinized and antigen retrieval was performed in heated citrate buffer (pH 6.0) and/or Tris-EDTA buffer (pH 9) for 30 min.
Respective stainings without primary antibodies were used as a negative control. Along with tissue arrays, serial sections of melanoma specimens and normal controls were stained to assess reproducibility. At equal Ab-concentrations, TSA-based visualization is expected to yield a higher number of positive cells as compared to conventional immunofluorescence. We therefore established TSA-based visualization of primary Abs on control tonsil tissue, the golden standard for lymphocyte antigen detection in pathology, and performed a comparison for each Ab to validated staining patterns in human tonsil (as to the Human Protein Atlas 79 , available from www.proteinatlas.org ). Thereafter, we balanced the signal through dilution of the primary Abs to obtain staining levels and cell frequencies comparable to conventional immunofluorescence staining. The dilution of the CD19 antibody was optimized to allowing for the detection of CD19 low plasma cell-like cells as compared to patterns and frequencies obtained by CD138+ pooled IgA/IgG+ stainings on human tonsil and melanoma 17, 19 . In multiplex stainings, single primary Ab stainings were run in parallel to control for false positive results through incomplete Ab-TSA complex-"stripping" and false negative results through antigen masking (by incubation with multiple primary Abs, "umbrella-effect"). Spillover effects were controlled for anti-CD20-Ab stainings on tonsil with different Opal fluorophores by signal detection in adjacent components/channels and thereafter for exposure time settings upon acquisition of multiplex-stained tissue sections.
Tissue imaging, spectral unmixing and phenotyping Multiplexed slides were scanned on a Vectra Multispectral Imaging System version 2 (Perkin Elmer) as described 77,78 . Briefly, a spectral library from spectral peaks emitted by each fluorophore from single stained slides was created with the inform Advanced Image Analysis software (InForm 2.4, PerkinElmer) and used for spectral unmixing of multispectral images allowing for identification of all markers of interest. Autofluorescence was determined on an unstained representative study sample. To subtype TAB in TMA, cells were phenotyped as (i) CD19 + CD20 -CD38 + CD138plasmablast-like, CD19 + CD20 -CD138 + plasma cell-like, CD19 + CD20 + CD38 -CD138memory B cell-like, CD20 + CD38 + CD138 -CD5germinal center B cell-like, CD19 + CD20 -CD38 -CD138 -CD27 + activated B cell-like, CD20 + CD19 -CD138 -CD5 + transitional cell-like TAB and (ii) other cells. The staining protocol has been optimized for detection of CD19. Though CD19 low plasma cell-like cells could be detected at significant numbers, they may still be underrepresented in our staining data. The same may also be true for the detection of activated B cell-like cells, as expression of CD27 has been reported to be downregulated on TAB 40,80 . Human melanoma cells can express CD38 81 and we observed some cores with slight CD38 immunoreactivity of melanoma cells. In these cores tumor cell areas were excluded with the "exclusion area tool" of InForm. After adaptive cell segmentation, a selection of around 25 representative original multispectral images was used to set cut-off values for each fluorophore/antibody staining. All phenotyping and subsequent quantifications were performed blinded to the sample identity.