Clinical data were collected from 57 patients from the RWTH Aachen University Hospital. All patients provided informed consent and analysis of human samples was approved by the local ethics board of the medical faculty RWTH Aachen (EK 206/09 and EK 391/20). MF grades were assessed by the local pathology department of the RWTH Aachen University Hospital for all patients except 3. Median age ± standard deviation of patients was 54.7 ± 15.4 years, 42 (74%) were male. Correlations between MF grade at diagnosis and TFR were assessed for patients undergoing discontinuation of the TKI after being at least 1.5 years in deep molecular remission (MR4 (Cross et al., 2012), BCR::ABL1 <0.01%). Success of discontinuation was defined as stable MMR for at least 1 year without need for re-treatment. TFR failure was defined as loss of MMR. Optimal response (OR) at 12 months was achieved if the patient’s BCR::ABL1 levels were below 0.1% (MMR as defined in the International Randomized Study of Interferon and STI571 trial [IRIS] (Hughes et al., 2010)) 12 months after diagnosis and start of TKI treatment.

FVB/N SCLtTA/BCR::ABL1 (double transgenic, CML) mice and FVB/N SCLtTA (single transgenic, control) siblings were bred in-house and tetracycline was added to the drinking water (0.5 g/L, Sigma-Aldrich, St. Louis, MO, United States) in order to avoid the expression of BCR::ABL1 (Institute for Laboratory Animal Science, Uniklinik RWTH Aachen). BM cells of control or CML mice were isolated from femur and tibia, separated by cell straining and 2 × 10⁶ cells were transplanted into the wildtype 10 Gy irradiated FVB/N CD45.1⁺ recipient mice by tail vain injection (Janvier Labs^®). Mice were treated with cotrimoxazole (Ratiopharm, Ulm, Germany) for 2 weeks after transplantation. One week after transplantation, BCR::ABL1 expression was induced by withdrawing tetracycline. After an additional week, treatment of CML mice with nilotinib (NIL, 50 mg/kg body weight) by daily gavage over 3 weeks was started. NIL (MedChemExpress, Monmouth Junction, NJ, United States) was resolved in 10% N-Methyl-2-pyrrolidon (Carl Roth) and 90% polyethylene glycol 300 (Sigma-Aldrich). Treatment controls were performed on both control and CML mice by administering the vehicle following the same procedure. All animal experiments were approved by the local authorities of North Rhine-Westphalia, Germany (LANUV Az. 84–02.04.2013.A072).

Formalin preserved sternum specimen were embedded in paraffin and sectioned with a microtome and further processed following manufacturer instructions (Cell Signaling^®). To perform the staining, sections were de-paraffinized and hydrated, followed by antigen retrieval with citrate buffer. Sections were then blocked by peroxidase followed by serum blocking. Sections were incubated overnight at 4°C with an anti-alpha-smooth muscle actin (α-SMA) primary antibody (add antibody number) at a 1:800 dilution. Next, slides were washed and incubated for 30 min with the secondary antibody (HRP, Rabbit #8114; Cell Signaling Technology^®). Finally, sections were developed with diaminobenzidine, counterstained with hematoxylin and mounted with coverslips. Quantification of IHC staining was carried out using Fiji software (ImageJ^®), after performing a color deconvolution (H DAB) (Crowe and Yue, 2023).

After Formalin preservation, decalcification (EDTA) and paraffin embedding, 2.5–4 µm slides were produced by a microtome. For staining, a silver impregnation technique was used. The sections were de-paraffinized and hydrated and workflow was performed according to the following protocol: potassium permanganate (1%) for 5 min, oxalid acid (2%) for 5 min, iron alum for 5 min, silver solution ammoniacal for 10 s, formalin (5%) for 3 min, sodium thiosulphate (2%) for 2 min. Between each step, washing with distilled water was performed. Afterwards, nuclear fast red solution was used for 10 min for counterstaining. After immersing in alcohol (ascending, 70%–100%) and xylol, the slides were mounted with coverslips.

Grading occurred according to the WHO grading system (semiquantitative bone marrow fibrosis (MF) grading system proposed by WHO Classification of Tumours of Hematopoetic and Lymphoid Tissues (Swerdlow et al., 2017)). The four grades (MF-0 to MF-3) are defined as follows: MF-0: Scattered linear reticulin with no intersections, MF-1: Loose network of reticulin with many intersections, especially in perivascular areas, MF-2: Diffuse and dense increase in reticulin with extensive intersections, occasionally with focal bundles of thick fibers mostly consistent with collagen and/or associated with focal osteosclerosis, MF-3: Diffuse and dense increase of reticulin with extensive intersections and coarse bundles of thick fibers, consistent with collagen, usually associated with osteosclerosis. In heterogenous patterns, the final grade is determined by the highest grade present in at least 30% of the marrow area.

BM and spleen cells were harvested and lysis of red blood cells was performed with a ammonium-chloride-potassium solution (AKC). Subsequently, cells were washed with PBS (Pan Biotech^®) + 2% fetal bovine serum (FBS, PAN Biotech^®) and then incubated for 20 min at 4°C with a 1:100 dilution with anti-CD41 conjugated with APC Cy7 fluorochrome (BioLegend^®) in the dark. Next, cells were washed and resuspended in PBS + 2% FBS. In addition, FACS staining of B-cells (B220⁺), granulocytes (GR-1⁺/CD11b⁺), T-cells (CD3⁺), blasts (KIT⁺), and erythrocytes (Ter119⁺) was performed. Flow cytometer measurements were carried out using Gallios flow cytometer (Beckman Coulter^®) and analyzed with Kaluza analysis software (Beckman Coulter^®).

Human MSCs were obtained from BM aspirates of five CML patients at time of diagnosis and from hip bone donated after hip replacement surgeries of five healthy donors (HD). Cells were isolated by tissue culture plastic adherence after the mononuclear cell fraction was separated by Ficoll. Cells were seeded at 160,000 cells per cm² in Dulbecco’s Modified Eagle Medium (DMEM) low glucose (Gibco^®) supplemented with 10% of FBS, 1% each of L-glutamine, penicillin/streptomycin (both Gibco^®) and cultured at 37°C with 5% CO₂. After 24 h, the medium was replaced and non-adherent cells removed by washing with phosphate-buffered saline (PBS, PAN Biotech^®). Afterwards, adherent cells were maintained with medium replacement twice a week, detached with 0.25% trypsin-EDTA (Gibco^®) after reaching confluence. MSCs were used for our experiments at passage 2 (p2).

MSCs were tested for their ability to differentiate in vitro into adipocytes, osteocytes and chondrocytes (Sudo et al., 2007). Briefly, in order to perform adipogenic differentiation, cells were cultured in medium consisting of DMEM high glucose (Gibco^®), 10% FBS, 1 µM dexamethasone, 200 µM indomethacin, 500 µM 3-isobutyl-1-methylxantine and 10 μg/mL insulin (all Sigma^®). On day 14–21, adipogenic cultures were washed with PBS, fixed in PFA 4% and stained with Oil Red O (Sigma^®). For the osteogenic differentiation, cells were cultured in medium consisting of DMEM high glucose (Gibco^®) supplemented with 10% of FBS (PAN Biotech^®), 1% L-glutamine, 1% penicillin/streptomycin (both Gibco^®), 10 nM dexamethasone, 10 mM β-glycerolphosphate and 50 µM ascorbic acid (Sigma^®). After 21 days, cells were washed with PBS, fixed in PFA 4% and stained with Alizarin Red (Sigma^®). For the chondrogenic differentiation, cells were cultured in medium consisting of DMEM high glucose (Gibco^®), 1% L-glutamine, 1% penicillin/streptomycin (both Gibco^®), 100 nM dexamethasone, 0.17 mM L-Ascorbic acid-2-PO4, 100 μg/mL sodium pyruvate, Insulin Transferrin Selenium supplement (ITS, Sigma^®) and 10 ng/mL TGF-β. On day 14, cultures were washed with PBS, fixed in PFA 4% and stained with Alcian blue (Sigma^®). After visual inspection and image acquisition, quantification of the three stainings was performed using the following solvents: isopropanol for adipogenic staining, 10%-cetylpyridinium chloride solution for osteogenic staining and 6M Guanidine HCl solution for chondrogenic staining, the relative absorbance was measured with a plate reader (Multiskan FC, Thermo Fisher^®).

MSC phenotype was assessed at passage 2 (p2) using the following monoclonal antibodies: anti-CD73-FITC (eBioscience^®); anti-CD34-FITC (BD Biosciences^®, BD); anti-CD45-PE (BD^®); anti-CD90-PerCP-Cy5.5 (eBioscience^®); anti-CD13-PE (eBioscience^®); anti-CD11b -FITC (BD^®) anti-CD44-APC and anti-HLA-DR-FITC (BD^®). Staining of at least 50,000 cells was performed at 4°C for 15 min in the dark in presence of 1% human serum to avoid unspecific binding. Samples were acquired with a Gallios flow cytometer (Beckman Coulter^®) and analyses were performed using FlowJo Software (BD^®).

Log serial dilutions of the MSC at p2 were placed into 10 to 20 replicates in 96-well plates in the presence of MesenPure (STEMCELL Technologies) and 0.5% penicillin-streptomycin. After 14 days, colony-forming unit-fibroblast (CFU-F) were performed using MesenCult Expansion Kit (STEMCELL Technologies^®), where the wells were stained with Giemsa (Merck) according to manufacturer’s instruction. Colonies were then counted manually. The frequency of progenitor cells in the undiluted starting population was calculated manually.

RNA of four HD- and four CML-derived MSCs was isolated using TRIzol™ Reagent (Thermo Fisher^®) as previously described. RNA was hybridized on Human Clariom D Assay gene chip (Affymetrix^®). Further steps, such as sample labeling, hybridization, washing and scanning were performed according to manufacturer’s protocols. Downstream analysis was performed in R (v4.2.2), using packages oligo (v1.62.2), Biobase (v2.58), affycoretools (v1,70), limma (v3.54.2), PCAtools (v2.10), progeny (v1.20), msigdbr (v7.5.1) and clusterProfiler (v4.6) for analysis; data.table (v1.14.8), tibble (v3.2.1) and dplyr (v1.1.1) for data wrangling; and ggplot2 (v3.4.1), ggrepel (v0.9.2), ComplexHeatmap (v2.14), patchwork (v1.1.2) and cowplot (v1.1.1) for visualization.

Robust Multichip Average preprocessing was performed. This included background subtraction, quantile normalization and summarization. The latter step was performed at the probeset level. Principal Component Analysis was performed to evaluate if each sample group was clustering together (Supplementary Figure S3B). Samples CML2 and HD4 were deemed unfit for this analysis. CML2 was treated with imatinib which could be a confounding factor. HD4 was clustering apart from the other HDs and therefore considered an outlier. Non-mapping probes, probes mapping to sexual chromosomes and probes with missing data were filtered out. A PCA was again performed after the filtering (Supplementary Figure S3C).

Differential expression analysis was performed with limma-trend with an exclusion filter for probes with a mean expression below 5 on the log2 scale, note that this still retains 439,961 probes. A linear model for the CML vs. HD comparison was fit and its statistics were estimated with empirical Bayes moderation. p-values were corrected for multiple testing with the Benjamini–Hochberg procedure.

PROGENy was performed on the same data fed to the differential expression analysis and setting the “top” parameter to 1,000 and “scale” to false. PROGENy results were then scaled in regard to the HD samples. Over-Representation Analysis (ORA) and Gene Set Expression Analysis (GSEA) were performed on the results from the differential expression analysis. For ORA, the background probes considered were the same as the ones present in the differential expression analysis. Only enrichment for under-expressed probes was performed as no probe passed the over-expression significance threshold during differential expression analysis. For ORA and GSEA, p-values were corrected for multiple testing with the Benjamini–Hochberg procedure. Chosen custom gene sets and MSigDB gensets were evaluated separately. For PROGENy and GSEA, probe information was summarized down by mean to the gene level.

1-way or 2-way analysis of variance (ANOVA) were applied for multiple comparison analysis, with Tukey’s or Bonferroni’s post-hoc-test. Unpaired t-test was used to compare the means of two groups. Fisher’s exact test was used for Contingency Table. Statistical calculations were performed with GraphPad Prism (GraphPad Software, La Jolla, CA, United States). Data are presented as mean ± standard deviation (SD). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 were considered statistically significant.

Based on the conflicting reports on the frequency of MF, we first assessed the proportion of patients with MF at diagnosis in our CML cohort (n = 57). Here, 38.6% (n = 22) had MF at the time of diagnosis, with 31.6% having MF stage 1 (n = 18) and 7.0% (n = 4) having stage 2 (Figure 1A). No grade 3 MF was observed. As MF has previously been suggested to be an adverse prognostic marker in CML (Buesche et al., 2003; Buesche et al., 2004; Buesche et al., 2007) we started to compare different clinical parameters among patients with different stages of MF. Here we observed that the percentage of peripheral blood (pB) and BM blasts is significantly higher in CML patients with MF grade 2 (Figures 1B,C). Spleen length, the number of BM megakaryocytes (MKs), platelet counts and leukocyte counts did not significantly change upon MF development (Figures 1D–G). In our cohort, presence of MF also does not correlate with response to treatment at 12 months (Supplementary Figure S1A), risk scores (Supplementary Figure S1B–D) or patient age at diagnosis (Supplementary Figure S1E). Furthermore, TFR outcome also did not correlate with common risk scores (Supplementary Figure S1F–H).

Using the transgenic CML model we previously confirmed that LSCs persist despite BCR::ABL1 inhibition (Schemionek et al., 2010; Hamilton et al., 2012). Further studies have revealed that CML development induces expansion of osteoblastic lineage cells and MF has been suggested to occur in these mice as analyzed by collagen (Schepers et al., 2013) or reticulin-staining (Reynaud et al., 2011). As the grade of fibrosis is usually low in CML, we here used α-SMA staining in order to dissect early fibrotic changes. To exclude any possibility of transgene expression in the non-hematopoietic compartment we first transplanted 2 × 10⁶ BM cells, isolated from either SCLtTA/BCR::ABL1 CML or SCLtTA control mice into 10 Gy irradiated recipients and allowed these cells to engraft for 1 week before inducing BCR::ABL1 expression. 2 weeks after induction of the oncogene, mice were treated with either nilotinib (NIL) or vehicle (V) control for 3 weeks (Figure 2A). Upon autopsy the increase in spleen weight was evident in the CML V group compared to the control V group. This was largely reverted upon NIL treatment as expected and previously shown (Figure 2B). In line with these data, expression of BCR::ABL1 was evident in V-treated CML mice but significantly decreased upon NIL treatment (Figure 2C). A detailed FACS immunophenotyping revealed that the B-cells were driving the disease phenotype and responded to TKI treatment (Supplementary Figure S2), as we had previously observed (Butow et al., 2023). Sternum sections of the BM were stained for α-SMA that is known to be expressed in myofibroblasts. Indeed, we observed a significant increase in V-treated CML mice that was reverted to some extend upon NIL treatment (Figures 2D,E). To our surprise, α-SMA appeared to be also expressed in CML MKs, which are a supposed driver of fibrotic transformation in MPN. FACS analysis revealed that MK percentages were not altered in the BM but increased in the spleen of CML transgenic mice and interestingly elevated in both organs upon TKI treatment (Figures 2F,G).

Mesenchymal stromal cells (MSCs) were either isolated from BM of CML patients or the hip bone of HDs via plastic adherence (Figure 3A) and their immunophenotype was analyzed at p2 (Figure 3B). The FACS analyses showed the expression of common MSC markers, such as CD73, CD90, CD14 and CD44. Typical hematopoietic markers, such as CD34, CD45, CD11b and HLA-DR were not present, consistent with the characteristic pattern of mesenchymal surface markers. Next, we evaluated their differentiation potential into the adipogenic, osteogenic, and chondrogenic lineage (Figure 3C). The adipogenic potential of CML-derived MSCs was significantly enhanced compared to HD. The osteogenic differentiation of CML-derived MSCs appeared to be elevated but that was rather due to a single CML sample showing an outstanding osteogenic capacity. Chondrogenic differentiation did not differ in CML vs. HD-derived samples. In order to evaluate the clonogenic capacity of CML MSCs, we performed a limiting dilution CFU-F assays (Figure 3D). These data show that there was no difference observed in CML-vs HD-derived MSCs. Overall, although CML-derived MSCs showed an increased capacity to differentiate into the adipogenic lineage, no other substantial phenotypic differences were observed in the two groups.

To determine the gene expression signature of CML-vs HD-MSCs, we performed a microarray analysis. Overall changes where rather mild in these MSCs that were expanded until p2, as shown by the volcano plot (Supplementary Figure S3A). However, initial quality control showed specific clustering of CML vs. HD samples and revealed HD 4 as an outlier (Supplementary Figure S3B), potentially due to technical reasons and was therefore excluded from subsequent analysis. In addition, CML 2 was previously treated and therefore likewise excluded from further analyses (Supplementary Figure S3C). We could observe significant changes in a defined fibrotic gene signature set, which was previously derived from the analysis of expression profiling datasets of human organs with fibrosis-related diseases (Wang et al., 2020) (Figure 4A). Furthermore, an altered expression profile was also observed in a defined gene set for hematopoiesis support (Figure 4B). Next, gene set expression analysis (GSEA) was performed using the molecular signature database (MSigDB) hallmark gene sets collection (Figure 4C). A significant alteration of signaling pathways involved in inflammation was found, such as interferon gamma and alpha response. Interestingly, a significant upregulation of epithelial-to-mesenchymal transition (EMT) signature was also observed. Among the downregulated pathways (Figure 4D), another gene set of high interest that was significantly downregulated was the one containing genes with at least one occurrence of the motif detected by MIR431 (Jiang et al., 2018). This microRNA has been previously found downregulated in AML and its overexpression induced downregulation of TGF-β as well as epithelial-to-mesenchymal transition-related genes. Finally, we show 14 additional signatures of pathway perturbation (pathway responsive genes, PROGENy analysis) (Schubert et al., 2018) in our two cohorts and found a significant upregulation of the TGF-β pathway in the CML-derived MSCs (Figure 4E).

The success of TFR is compromised by the persistence of LSCs in CML patients on TKI therapy, which re-expand upon TKI discontinuation. Based on the expression analysis of primary patient-derived MSCs indicating LSC-supporting potential, we investigated the impact of MF on TFR. A minority of CML patients underwent TKI discontinuation. In our cohort of patients with available MF grading at diagnosis, 12 patients discontinued TKI therapy after at least 1.5 years of MR4 or better. The clinical parameters of these patients are described in Table 1 and Supplementary Table S1. In the small cohort of patients, no significant differences in age, treatment duration or duration of MR4/5 were observed when comparing patients who succeeded or failed TFR. Of those 12 patients, 4 failed TFR (33%) and interestingly, all presented with MF at diagnosis (Figure 5). There was one additional patient with MF grade 1 at diagnosis who failed TFR, but he was excluded from this analysis, as he only reached MR3 for 16 months and therefore did not fulfil the current criteria for TKI discontinuation. Out of 8 patients who achieved TFR, 2 (25%) had MF grade 1 at diagnosis, whereas the other 6 (75%) had no BM fibrosis. Notably, all patients without MF at diagnosis achieved TFR, while all patients with MF grade 2 at diagnosis, failed. Only patients with MF grade 1 varied in TFR success, half of them failing, the other succeeding (Table 1).