A complex interaction between blasts and surrounding cells in the acute myeloid leukemia (AML) bone marrow (BM) microenvironment sustains blast proliferation and confers chemoresistance. T- and NK-cells have been shown to be dysfunctional in AML, which might be associated with immune evasion and poor prognosis. Here, we present a comprehensive analysis of the immune contexture of the AML BM at diagnosis and study its interaction with clinicopathological variables.


Diagnostic BM biopsies (n=69) were collected from AML patients treated in the Helsinki University Hospital during 2005-2017 and age and gender-matched controls (n=12) to construct tissue microarrays (TMA). Using 8-plex immunohistochemistry (mIHC) and computerized image analysis, we determined cell abundance and immunophenotypic states of millions of immune cells. Immunoprofiles were integrated with a total of 120 clinicopathological variables including cytogenetics and molecular genetics, ELN (European Leukemia Net) risk classification, disease burden parameters, and patient demographics.


Unsupervised hierarchical clustering of the immunologic contexture defined by mIHC analysis grouped AML patients distinctly from control subjects (Fig 1a). By extracting significant differences (Mann-Whitney U test, q<0.05) and annotating immunologic markers as either anti-cancer or immunosuppression drivers based on literature, we observed an interesting polarization of increased immunosuppression in AML compared to control BM (Fig 1b). In AML patients, lower fraction of granzyme B expressing (GrB+) cells was noted both in CD3+CD4+ helper T-cells (10.9% vs 24.3%, q=0.002, in AML vs control BM) and CD3+CD8+ cytotoxic T-cells (23.5% vs 33.5%, q=0.02). Moreover, we observed pronounced T-cell inhibition features, such as higher proportion of regulatory T-cells (1.5% vs 0.0% FOXP3+ of helper T-cells, q<0.001), and lower expression of class I HLA in BM cells (89.1% vs 100.0%, q<0.001). Putative exhausted PD1+ T-cells were also markedly enriched in the AML BM (15.7% vs 1.7% PD1+ of helper T-cells and 13.2% vs 2.0% PD1+ of cytotoxic T-cells, q<0.001).

Among the high interpatient heterogeneity, we discovered two main immune profiles. Cluster 1 was characterized with higher proportion (Log2 fold change >0.5, q<0.05) of cytotoxic T-cells and expression of CD57, CD27, and CD25 in T-cells, as well as higher expression of PD-L1 in the BM. Moreover, lower expression (Log2 fold change <0.5, q<0.05) of OX40 and CD45RO in T-cells and proportion of CD11c+BDCA1+ (type 1 myeloid dendritic cells) were observed. Patients of Cluster 1 were associated with longer event-free survival (EFS; HR 1.9, p=0.049) as well as lower age (median 53.6 vs 64.4 years, p=0.001). No connection between immunologic clusters and FLT3 or NPM1 genotype, complex karyotype, ELN risk class, blast proportion or leukocyte count was found.

To support clinical decision-making in ELN 2017 intermediate-risk patients, we developed a risk stratification model focusing on this particular subgroup (n=28) using L1-penalized Cox regression. Patient age over 60 years (HR 8.1, CI95% 2.5-26.6 p<0.001) and low proportion of CD45+CD2+CD3- NK-cells (HR 0.92, CI95% 0.85-0.99 p=0.03) predicted worse EFS.

In intermediate-risk patients (n=68) of a separate validation cohort (n=145) analyzed with flow cytometry, low NK cell proportion and high age predicted worse EFS (HR 2.7 CI95% 1.6-4.6 p<0.001) and OS (HR 3.9 CI95% 2.1-7.3 p<0.001) after adjusting with induction therapy protocol. Lower NK-cell proportion was associated with FLT3-ITD genotype (0.45% vs 1.1% NK-cells/all cells in FLT3-ITD+ vs FLT3-ITD- AML patients, p=0.01), and higher than median PB leukocyte (WBC) count (0.50% vs 1.9% NK-cells/all cells in patients with PB WBC ≥7.8x10E9/L vs <7.8x10E9/L, p<0.001).


Using TMA cytometrics with mIHC and automated image analysis for detailed characterization of the immune contexture, we discovered pronounced immune exhaustion and suppression in AML BM. An aging-related immune profile was identified and was associated with poor prognosis. Furthermore, survival prediction of intermediate-risk patients might be enhanced by considering patient age and NK-cell proportion. Taken together, immunophenotyping of AML patients might improve risk stratification and identify a subgroup benefiting from immunomodulatory treatments.


Pallaud:Novartis: Employment. Marques Ramos:Novartis: Employment. Porkka:Novartis: Honoraria, Research Funding; Celgene: Honoraria, Research Funding. Mustjoki:Ariad: Research Funding; Novartis: Honoraria, Research Funding; Celgene: Honoraria; Bristol-Myers Squibb: Honoraria, Research Funding; Pfizer: Honoraria, Research Funding.

Author notes


Asterisk with author names denotes non-ASH members.