Formalin-fixed and paraffin-embedded (FFPE) MS samples and bone marrow biopsies (BMB) (n=83) from 45 patients (38 MS samples with paired BMBs and further 7 MS cases without paired BMB) were collected in the period from 2011 to 2022 and archived at the Institutes of Pathology of the Martin-Luther University Halle-Wittenberg, Germany (n=29) and the University of Leipzig, Germany (n=16), respectively. The use of the FFPE BMB was approved by the Ethical Committee of the Medical Faculty in Halle, Germany (2017-81 and 2023-196). Clinical data from these patients were available, such as age, gender, disease status, therapy and survival time (see Table 1 ).

Histopathological diagnostics were performed according to the diagnostic criteria of the World Health Organization (WHO) classification of Tumors of Hematopoietic and Lymphoid tissues, fourth edition 2017 and 2022 (3, 4, 19). Conventional histopathology of the MS cases was performed employing H&E staining and chloroacetate esterase reaction. Immunohistochemistry (IHC) was performed on all samples using antibodies (Ab) directed against CD33, CD34, CD117, MPO, lysozyme, CD68, HLA-I HC, ß₂M, tapasin (tpn), TAP1, TAP2 and HLA-G according to the suppliers’ instructions. The Ab are summarized in Supplementary Table S1 . For the expression analysis of HLA-I HC, ß₂M, tapasin (tpn), TAP1, TAP2 and HLA-G, the H score was employed as described elsewhere (20). A high expression of the respective proteins refers to a H score >150.

In order to compare the APM component expression in human AML cells that showed mass formation in vitro and in vivo, a publicly available dataset (GSE103344) (21) containing Affymetrix Human Gene 2.0 ST mRNA Array data of human THP-1 AML cells with knock down of RKIP that showed a role in tumor mass formation in vitro and in vivo (21). The differentially gene expression (DGE) of various APM components was analyzed using the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/). Differentially expressed genes (DEG) of AML cells with knock down of RKIP and respective controls with preserved RKIP were analyzed and visualized using the GEO2R tool (https://www.ncbi.nlm.nih.gov/geo/geo2r/).

Multispectral imaging (MSI) was performed as recently described (22) employing a six-plex Ab panel with CD3, CD8, FoxP3, MUM1p, CD34, and granzyme B (GrB). Briefly, after antigen retrieval at pH 6 or 9 depending on the Ab used, the tissues were incubated for 30 min with the primary Ab followed by the secondary Ab (Akoya biosciences, Marlborough, MA, USA, Opal Polymer HRP Ms + Rb) for 10 min. Tyramide signal amplification (TSA) visualization was performed using the Opal seven-color IHC kit containing the fluorophores Opal 520, Opal 540, Opal 570, Opal 620, Opal 650, Opal 690 (Akoya biosciences, Marlborough, MA, USA) and DAPI. Stained slides were imaged employing the Pheno Imager HT (Akoya biosciences, USA). Cell segmentation and phenotyping of cell subpopulations were performed using the inForm software (Akoya biosciences, USA). The frequency of immune cell populations and their cartographic coordinates were evaluated for immune cell enumeration and relationship analysis using the R scripts from the phenoptr and phenoptrReports packages (https://github.com/akoyabio). Moreover, all CD3 stains of the MS samples were analyzed by a pathologist and the immune cell infiltration pattern was grouped in “immune cell excluded” tumors with immune cells located at the margin of the tumor and “immune cell infiltrated” tumors with a diffuse immune cell infiltration into the tumor (23).

Four to five 10 µm thick FFPE tissue slides were prepared. Total RNA was isolated with Maxwell RSC RNA FFPE Kit (Promega, USA) according to the manufacturer’s instructions. For RNA sequencing (RNAseq), 2 μg of total RNA/sample was employed and strand specific 150 bases paired-end RNAseq was done using the Illumina NovaSeq platform by Genewiz (Leipzig, Germany). Approximately 20 million reads per sample were obtained. Reads were trimmed using Trimmomatic, quality checked using fastqc 0.11.9 and mapped to the human reference genome (hg38) using splice aware aligner STAR 2.7.9a. Quantifications on NCBI gene models (hg38refGene) were performed using featureCounts v2.0.0. Differential gene expression analyses between HLA-I HC^high versus HLA-I HC^low samples was performed based on a negative binomial distribution using the R package DESeq2 (24). P-values were adjusted based on the false discovery rate (FDR) according to the Benjamini-Hochberg method. Genes with an average expression across all samples (base mean) >10, more then two-fold change, and a FDR<0.1 were considered as significantly differentially expressed and visualized in a volcano plot using the R package EnhancedVolcano. Gene set enrichment analysis was performed on log2-fold changes with the GSEA tool v4.2.3 (25) using hallmark genesets (MSigDB) and results were visualized as bubble plots.

The Mann–Whitney U test was employed to compare clinical data. Patients with missing information in any other variable were excluded from regression analyses. Cox regression analyses were performed using IBM SPSS. Kolmogorov–Smirnov test revealed non-parametric data (p < 0.05). The Mann–Whitney U test was used to compare clinical data, frequencies of immune cell subpopulations and the expression pattern of immune-relevant markers. Survival analyses were performed on 24 patients (follow-up time of 25 months) using the Kaplan-Meier estimators and differences calculated with log-rank tests or Cox regression models. P values < 0.05 were considered statistically significant. The figures were generated using the GraphPad Prism 7.0 software, IBM SPSS Statistics 28.0 and biorender (https://www.biorender.com).

In order to determine the expression levels of HLA-I APM components in MS, protein expression of HLA-I HC, ß₂M, TAP1, TAP2 and tpn was determined by conventional IHC ( Figures 1A, B ). Low HLA-I HC and β₂m expression levels (H-score <150) were found in 50.0% (19/38 MS cases) and 63% (28/45 MS cases), respectively. Tpn expression was low in 57.7% (26/45 cases), TAP1 and/or 2 in 62.2% of MS samples (28/45 cases), respectively. In general, 37/45 MS cases (82.2%) showed an impaired expression of at least one APM component. As an alternative immune escape mechanism a high HLA-G expression was found in 12/44 cases (27.3%), which was accompanied in 9/12 (75.0%) cases by concordant high HLA-I HC expression levels with an H-score >150. No significant differences in results were detected by comparing tissue specimens of the two different pathology departments ( Supplementary Table S2 ). In addition, a publicly available dataset (GSE103344) of RNA expression analysis of human AML cell lines was evaluated for DGE. As shown in Figures 1C, D , a downregulation of HLA-I APM components with particularly decreased mRNA expression of the HLA-I HC and TAP1 was detected.

To elucidate the immune escape mechanisms within the evolution of MS in an individual patient, the HLA-I APM component expression was analyzed between matched BMB and MS tumors available from 38/45 MS cases (84.4%) ( Figure 2 ; Table 2 ). In 4 cases, no neoplasia in the bone marrow could be detected. A downregulation of HLA-I HC and ß₂M expression was found in 65.7% (25/38) and 63.1% (24/38) of MS tumor cases, respectively, while in 4 patients a higher HLA-I HC expression was detected in the MS vs. BMB. A downregulation of TAP1 was found in 40.5% of MS cases compared to corresponding BMB, while tpn expression was only downregulated in 27.1%, but upregulated in 29.7% of MS cases. Furthermore, HLA-G expression was higher in 27.0% of MS cases, but lower in 8.1% compared to BMB. No significant differences in the HLA-I APM component expression was found regarding the underlying diseases or the anatomical localization ( Supplementary Tables S3 , S4 ).

Since the immune cell infiltration of the TME from BM and paired MS might differ, the frequencies and the spatial distribution of CD3⁺CD8^- T cells, CD3⁺CD8⁺ T cells, CD3⁺FoxP3⁺ regulatory T cells (Treg) and CD3^-MUM1⁺ B cells/plasma cells, CD3⁺GrB⁺ T cells were analyzed in 38 paired BMB and MS samples. As representatively shown in Figures 3A, B , significant differences in the composition of the immune cell subpopulations and their localization were demonstrated between BMB and MS. In general, all analyzed immune cell subsets showed a lower mean frequency in MS cases. This was accompanied by lower numbers of GrB⁺ cells in MS suggesting an impaired T cell activity ( Figure 3C ). Analysis of the spatial distribution revealed a distinct pattern of immune cell infiltration in MS compared to BMB ( Figures 3D, E ). In BM, a diffuse infiltration and distribution of TIL was detected, while TIL were predominantly located in the periphery of the tumor tissue and in the proximity of blood vessels in most MS representing an immune cell excluded “cold” TME (32/45). Of note, MS cases with a “hot” TME characterized by high TIL numbers within the tumor formation were found in 3/4 isolated MS cases lacking BM pathology. The mean minimal distance of CD3⁺ and CD3⁺CD8⁺ T cells in the MS was higher when compared to the BM (6.38 µm versus 10.01 µm), but exhibited a high variability ranging from 2.31 - 92.23 µm ( Figure 3F ). Since the expression of immune-relevant molecules could be influenced by the immune cell repertoire and vice versa (26, 27), the interrelationship between the HLA-I APM component expression and the local immune cell infiltration was analyzed in MS. While the correlation of the total numbers of all TIL subsets analyzed with HLA-I HC expression demonstrated no significant difference, HLA-I^high cases (H score >150) had a higher frequency of CD3⁺CD8⁺ T cells as well as higher numbers of GrB⁺ T cells ( Figure 3G ).

Based on the interrelation between immune-relevant molecules and the TME composition, the clinical relevance of the HLA-I APM component expression and immune cell infiltration was determined in the MS patients. The average OS of the 24 MS patients with available outcome (follow up time of up to 25 months) was 7.8 months. Forrest plot depiction of univariate cox regression demonstrated a significant influence of HLA-I expression in MS, but not in the corresponding BM. Moreover, higher HLA-I and ß₂M expression levels accompanied by increased TIL numbers correlated tendentially with a better patients’ outcome ( Figure 4A ), which is also underlined by Kaplan Meier estimators for TIL, T cell numbers and HLA-I HC ( Figure 4B ).

To get insights into the underlying cause of the better clinical outcome of patients with HLA-I^high tumors, the transcriptome of 5 MS cases with high/preserved (HLA-I^high) and 5 MS cases with reduced HLA-I (HLA-I^low) expression was determined by RNAseq analyses. Principal component analysis (PCA) revealed that 7/10 MS samples were grouped together. The variation showed neither an association with the underlying BM findings nor with the HLA-I HC expression ( Figure 5A ). As shown in a volcano plot, DGE analysis revealed 93 significantly upregulated and 57 significantly downregulated genes (fold change >2, p<0.05) in HLA-I^high versus HLA-I^low samples, respectively ( Figure 5B ). The 10 most upregulated genes were involved in immune signaling metabolism and cell differentiation and include e.g. DNTT, PROM1, and FCRL1, while the 10 most significantly downregulated genes in HLA-I^high cases were transcription factors and genes involved in immune or cell signaling and/or exhibit enzymatic activity ( Table 3 ). Moreover, gene set enrichment analysis (GSEA) ( Figure 5C , for all pathways see Supplementary Figure S1 ) revealed in samples with preserved HLA-I HC expression a downregulation of inflammatory response genes with significantly lower expression levels of genes involved in TNF-α signaling and interferon-γ response when compared to HLA-I^low samples. Moreover, a decreased expression of E2F, the MYC-targets V1 and V2, cell cycle checkpoints and metabolic pathway components was shown in HLA-I^high cases. A sustained interferon-γ response was also found in MS samples with high TILs when compared with patients with low TILs in MS formation ( Supplementary Figure S1 ). The RNAseq data compared to the GSE103344 dataset showed a down-regulation of HOXB9 and an up-regulation of CTSG in cells that showed mass formation in vitro and in vivo ( Supplementary Table S2 ).