Article Text

Download PDFPDF

108 Unveiling the diversity in melanoma immunotherapy response biomarkers between lymph node and non-lymph node biopsies
  1. Ettai Markovits1,
  2. Nethanel Asher2,
  3. Guy Ben-Betzalel3,
  4. Assaf Debby4,
  5. Becky Arbiv1,
  6. Ron Elran1,
  7. Yuval Shachaf1,
  8. Shai Bookstein1,
  9. Lena Tsabari1,
  10. Pinchas Birnbaum1,
  11. Tal Dankovich1,
  12. Amit Bart1,
  13. Yoad Cohen1,
  14. Kenneth Bloom1,
  15. Oscar Puig1,
  16. Ronnie Shapira-Frommer5 and
  17. Iris Barshack6
  1. 1Nucleai, Tel Aviv, Israel
  2. 2Sheba Medical Center, Ella Lemelbaum Institute of Immuno-Oncology, Israel
  3. 3Ella Lemelbaum Institute for Immuno-Oncology, Ramat Gan, Israel
  4. 4Sheba Medical Center, Ramat Gan, Israel
  5. 5Sheba Medical Center, Tel Aviv, Israel
  6. 6Tel-Aviv University, Tel Aviv, Israel
  • Journal for ImmunoTherapy of Cancer (JITC) preprint. The copyright holder for this preprint are the authors/funders, who have granted JITC permission to display the preprint. All rights reserved. No reuse allowed without permission.


Background Immunotherapy revolutionized the treatment of melanoma, however, a significant proportion of patients fail to respond. Thus, reliable biomarkers are needed to predict treatment response. Spatial biology, which investigates the spatial distribution and interactions of cells within tissues, has emerged as a powerful approach to unravel the complexity of the tumor-microenvironment (TME). In this study, we analyzed multiplex immunofluorescence (mIF) images from melanoma patients treated with immunotherapy, aiming to identify spatially resolved biomarkers associated with response.

Methods Two consecutive slides from pretreatment biopsies of 47 melanoma patients who received PD-1 blockade, as a single-agent, or combined with ipilimumab, were collected and stained with two 6-plex mIF antibody panels using PhenoImager, followed by hematoxylin and eosin (H&E) stain on the same slide (figure 1). Utilizing our deep learning-based multiplex analysis pipeline1 we were able to identify 9 distinct cell types across the two panels (figure 2). These predictions were evaluated against expert annotations. Corresponding sections stained with H&E provided additional cellular information such as the identification of fibroblasts and granulocytes, and tumor or TME area region assignment. More than 1,300 spatial features were calculated using a combination of cell type, marker positivity and area assignments, as well as cell interactions, and were compared between responders and non-responders to immunotherapy using Welch’s t-test in lymph node (LN) and non-LN tumor biopsies.

Results Our analysis pipeline demonstrated excellent performance both in accurately defining cell types and single-markers positivity (figures 1 and 2). As the TME is vastly different in LN and non-LN biopsies, we aimed to characterize different spatial features which are associated with response to treatment in each biopsy type. In LN biopsies, we observed CD8+ FOXP3+ T-cells upregulation in the TME of responders to PD-1 blockade, while fibroblast and granulocyte interactions in the TME correlated with poor response (figure 3). However, these associations were not observed in non-LN biopsies, where a high number of B-cell and T-cell interactions in the tumor invasive margin, and tumor cells and lymphocyte interactions in the tumor core were associated with treatment response. Additionally, a high number of granulocytes in the TME of non-LN biopsies was associated with treatment resistance (figure 4).

Conclusions Our study demonstrates that the tumor-microenvironment in LN and non-LN melanoma biopsies exhibits distinct spatial characteristics associated with immunotherapy response. These findings highlight the importance of spatial features, cell composition, and biopsy site when identifying spatial biomarkers for immunotherapy response prediction.


  1. Markovits E, Dankovich T, Gluskin R, et al. A novel deep learning pipeline for cell typing and phenotypic marker quantification in multiplex imaging. bioRxiv, 2022.

Ethics Approval This single-center, retrospective study of medical records was approved by the Institutional Review Board of the Sheba Medical Center (4387–17-SMC).

Abstract 108 Figure 1

Deep learning-based multiplex imaging analysis pipeline performance in single-markers positivity classification in panel 1 (A) and panel 2 (B).

Abstract 108 Figure 2

Deep learning-based multiplex imaging analysis pipeline performance in cell typing in panel 1 (A) and panel 2 (B).

Abstract 108 Figure 3

(A) The top differentially expressed features in between responders (n=7) and non-responders (n=6) to PD-I blockade in lymph node biopsies and their distribution in responders (n=19) and non-responders (n=15) in non-lymph node biopsies. (B) A representative image of a tumor-microenvironment enriched with CD8+ FOXP3+ T-cells (white arrows).

Abstract 108 Figure 4

(A) The top differentially expressed features in between responders (n=19) and non-responders (n=15) to PD-1 blockade in non-lymph node biopsies, and their distribution in responders (n=7) and non-responders (n=6) in lymph node biopsies. (B) A representative image of a tumor infiltrating T-cells.

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.