henhao
Sign in
Cancer BiologyScience Advances
Peer-reviewed journal article

Single Cell Spatial Transcriptomics Reveals Immunotherapy-Driven Bone Marrow Niche Remodeling in AML

Gege GuiiD, Molly A. BinghamiD, Julius R. Herzog, Abigail Wong-RolleiD, Laura Williams DilloniD, Meghali GoswamiiD, Eddie Martin, Jason ReevesiD, Sean T. KimiD, Arya J. Bahrami, Hermann Franz DegenhardtiD, George ZakiiD, Prajan DivakariD, Edward C. SchromiD, Katherine R. CalvoiD, Christopher Simon HouriganiD, Kasper D. HanseniD, Chen Ning ZhaoiD

PDFDOIPreprintCodeData
Post to X
Threads on XPost to BlueskyThreads on Bluesky
In-depth full-text digest available · Jump to digest ↓

Abstract

Given the successful graft-versus-leukemia cell treatment effect observed with allogeneic hematopoietic stem cell transplant for patients with refractory or relapsed acute myeloid leukemia, immunotherapies have also been investigated in the nontransplant setting. Here, we use a multi-omic approach to investigate spatiotemporal interactions in the bone marrow niche between leukemia cells and immune cells in patients with refractory or relapsed acute myeloid leukemia treated with a combination of the immune checkpoint inhibitor pembrolizumab and hypomethylating agent decitabine. We derived precise segmentation data by extensively training nuclear and membrane cell segmentation models, which enabled accurate transcript assignment and deep learning-feature-based image analysis. To overcome read-depth limitations, we integrated the single-cell RNA sequencing data with single-cell-resolution spatial transcriptomic data from the same sample. Quantifying cell-cell distances between cell edges rather than cell centroids allowed us to conduct a more accurate downstream analysis of the tumor microenvironment, revealing that multiple cell types of interest had global enrichment or local enrichment proximal to leukemia cells after pembrolizumab treatment, which could be associated with their clinical responses. Furthermore, ligand-receptor analysis indicated a potential increase in TWEAK signaling between leukemia cells and immune cells after pembrolizumab treatment. HighlightsO_LISpatial transcriptomic analysis of R-AML bone marrow niches provides detailed information about intercellular interactions in the tumor microenvironment. C_LIO_LIImmunotherapy shifts the cell composition of the leukemia neighborhood. C_LI Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=199 SRC="FIGDIR/small/634753v1_ufig1.gif" ALT="Figure 1"> View larger version (93K): org.highwire.dtl.DTLVardef@955273org.highwire.dtl.DTLVardef@158234corg.highwire.dtl.DTLVardef@175837borg.highwire.dtl.DTLVardef@b0a38_HPS_FORMAT_FIGEXP M_FIG C_FIG

02 /Specification & Classification

Click any tag to browse matching papers
Citations
26
FWCI
7.77
Field percentile
top 1%
OpenAlex Topics

In-depth summary

✦ Multi-agent verified full-text synthesis

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (27 Jan 2025) · CC0 1.0

CosMx spatial transcriptomics of bone marrow from 6 refractory/relapsed AML patients shows CD14+ monocytes rising in leukemia-cell neighborhoods after pembrolizumab in responders, dendritic cells in nonresponders.

Earlier transcriptomic profiling of T cells from this pembrolizumab plus decitabine trial in refractory/relapsed AML did not explain why patients responded differently, and dissociated scRNA-seq discards bone marrow spatial context. The work pairs scRNA-seq with antibody-derived tags and CosMx single-cell spatial transcriptomics, adding retrained segmentation plus neighborhood and density-shift analyses to follow niche remodeling across treatment.

Key results

  1. SuggestedAfter the immune checkpoint inhibitor (ICI) pembrolizumab, leukemia-cell neighborhoods shifted by response group: CD14+ monocytes rose in responders and dendritic cells (DCs) rose in nonresponders, uniformly across all distance rings examined.

    Poisson generalized linear mixed models (field-of-view random effect, Bonferroni) over rings 0-5 to 35-45 µm: responders (n=4) CD14+ monocytes rate ratio (RR) 2.5-2.7, adjusted p < 0.001, not significant in nonresponders; nonresponders (n=2) DCs RR 1.95-2.01, adjusted p < 0.001, opposite nonsignificant trend in responders. · Representing Leukemia Cell Neighborhoods with…; Generalized Linear Mixed Effect Model for…; Figure legends

  2. SuggestedA permutation-based density-shift method found patient-specific changes in which cell types lie near leukemia cells between time points, and some of these shifts were shared by patients with different clinical responses.

    Shift deemed significant if beyond 1 standard deviation of 100 permutation backgrounds. Naïve CD4+ T cells were denser near leukemia at baseline than end of cycle 2 in patients 2 and 6 (responders) and 3 (nonresponder); patient 3 also had baseline enrichment of DCs, monocyte progenitors, CD14 and CD16 monocytes; patients 2 and 4 gained monocyte progenitors. · Changes in Patient-Specific Spatial Cell…; Cell Type Density Shift; Figure legends

  3. SuggestedIn nonresponder patient 3 after ICI, leukemia cells expressed the TWEAK ligand gene TNFSF12, and naïve CD4 T cells and DCs within 5 microns expressed more of its receptor TNFRSF12A than same-type cells farther away. This suggests possible leukemia-driven TWEAK signaling.

    Pseudo-bulk Mann-Whitney U tests (FOVs as replicates) comparing cells within 5 microns vs at least 30 microns of leukemia, using 719 of 3,399 CellTalkDB pairs on the panel: TNFRSF12A higher near leukemia in naïve CD4 T cells and DCs (p < 0.05), monocyte progenitors p=0.07; all significant pairs were post-ICI. · Ligand-Receptor Analysis Reveals Leukemia to…; Ligand-Receptor Analysis; Figure legends

  4. ShownCustom-trained Cellpose 2.0 nucleus and membrane models were merged into one mask per cell, with nuclear expansion used only for cells lacking membrane signal. They segmented CosMx marrow images more accurately than the vendor default and better reflected cell size diversity.

    Trained on 448,439 nuclei and 318,401 membrane annotations; at IOU 0.7, F1 0.57 vs 0.23 for NanoString's default segmentation; 629,051 vs 593,279 cells detected across six patients; median cell size 46.1 µm vs 58.3 µm. · Further Training of Combined Nucleus-Cell…; Model Training; Implementing Segmentation Models

  5. ShownscRNA-seq with antibody tags found patient-specific leukemia populations that matched clinical immunophenotypes and cytogenetics. It also supplied marker references that, together with CD34 image features, were used to type CosMx cells in situ.

    Integrated 10x data: 55,635 cells; 5,468 leukemia cells (13.8% of patient cells), 91% overlapping per-patient malignant clusters; inferCNV in patient 3 recovered chr5/chr7 loss and chr21 gain. CosMx: 40,112 leukemia cells from CD34-channel EfficientNet features; 98% of cells larger than 648 µm labeled megakaryocytes. · Single-Cell Transcriptomic Analysis of Bone…; Cell Typing in situ by Protein Expression…; Cell Type Identification for Spatial…

Methods, data and reproducibility

Baseline marrow aspirates profiled by 10x scRNA-seq with antibody-derived tags supplied marker references for 960-gene CosMx imaging of biopsies at three time points. Cells were segmented with retrained Cellpose models, typed using CNN image features and InSituType, then analyzed with Poisson mixed models, permutation-based density shifts and ligand-receptor tests.

  • Baseline bone marrow aspirates from the pembrolizumab plus decitabine R-AML trial (NCT02996474), profiled by 10x 3' scRNA-seq with antibody-derived tags · human · n = 10 patients; 55,635 cells incl. 3 donors · GEO GSE271406
  • Healthy donor bone marrow used as the scRNA-seq integration and annotation reference · human · n = 3 donors
  • FFPE bone marrow biopsies at baseline, post-ICI (C1D8) and end of cycle 2, imaged as fields of view (FOVs) on a pre-commercial CosMx SMI with a 960-probe panel · human · n = 6 patients; 141 FOVs; 629,051 segmented cells · Zenodo 10.5281/zenodo.12730053
  • Annotated nucleus and membrane masks used to train and evaluate the Cellpose segmentation models · human · n = 448,439 nuclei; 318,401 membranes · Zenodo 10.5281/zenodo.12730053
  • CellTalkDB human ligand-receptor gene pairs · human · n = 3,399 pairs (719 on spatial panel)
Methods & tools
CosMx Spatial Molecular Imager (NanoString), 10x Genomics Chromium 3' v3 with TotalSeq-A antibody tags, Cell Ranger 3.1.0, Seurat v3, InferCNV, Cellpose 2.0 (custom-trained nucleus and membrane models), Huygens Professional 22.04, EfficientNet (B0 layer features), Phenograph, InSituType, Poisson generalized linear mixed model, CellTalkDB
Code
https://github.com/chenzhaolab2023/AML-Spatial
Data & accessions
https://zenodo.org/doi/10.5281/zenodo.12730053
Reagents & environment
R 4.3.2; Python 3 (minor version varied by virtual environment); Cell Ranger 3.1.0; Seurat version 3; Cellpose 2.0; Huygens Professional 22.04.
Reproducibility gaps
The CosMx images with annotations were to be posted on Zenodo only upon acceptance; patient clinical data appear only in the earlier trial paper; Python minor versions are unspecified; imaging used a pre-commercial CosMx instrument.
Limitations (6)
  • The CosMx 1000-plex panel detects at most 960 transcripts, far fewer than scRNA-seq, so some tumor-immune interactions are missed. Sparse per-cell transcript counts also made cell typing difficult without an scRNA-seq reference. · stated by the authors
  • Requiring CD34+ leukemia restricted spatial profiling to 6 of the 10 trial patients (n=4 responders, n=2 nonresponders). Patient-to-patient heterogeneity is hard to separate from treatment effects, and a larger trial is needed for validation. · stated by the authors
  • Needle biopsies give a narrow view, and only 5 to 15 FOVs per biopsy were imaged. Variability between same-sample FOVs suggests the chosen fields may not represent the whole specimen, and extra reporter cycling risks tissue detachment. · stated by the authors
  • The Poisson mixed model assumes uniform changes within each response group, uses coarse distance bins and ignores overlapping leukemia-cell neighborhoods. This may hide patient-specific signals. · stated by the authors
  • Patient-specific density shifts count as significant if they fall outside 1 standard deviation of only 100 permutations, with no stated multiple-testing correction across cell-type pairs and patients, so these patterns are exploratory. · noted in this summary
  • Some reported details conflict: the density-shift comparison is end of cycle 2 in Results but post-ICI in Methods and Figure 4; the ADT panel is 77 tags vs 32 antibodies; FOV totals are 141 vs 142. · noted in this summary

Who should read:Read this if you analyze imaging-based spatial transcriptomics of bone marrow or study immune checkpoint response in AML and want a workflow for segmentation, scRNA-seq-guided cell typing and leukemia-neighborhood analysis.

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISMeulen '26Dasdemir '25Ly '25Ennis '22Maurer '25Tian '26
69% semantic matchCancer Biology
Merel van der Meulen, Emma S. Pool, Alicia Perzolli+21 more·bioRxiv·Aug 7, 2026