<div>Abstract<p>Bone marrow trephine biopsy is crucial for the diagnosis of multiple myeloma. However, the complexity of bone marrow cellular, morphologic, and spatial architecture preserved in trephine samples hinders comprehensive evaluation. To dissect the diverse cellular communities and mosaic tissue habitats, we developed a superpixel-inspired deep learning method (MoSaicNet) that adapts to complex tissue architectures and a cell imbalance aware deep learning pipeline (AwareNet) to enable accurate detection and classification of rare cell types in multiplex immunohistochemistry images. MoSaicNet and AwareNet achieved an AUC of >0.98 for tissue and cellular classification on separate test datasets. Application of MoSaicNet and AwareNet enabled investigation of bone heterogeneity and thickness as well as spatial histology analysis of bone marrow trephine samples from monoclonal gammopathies of undetermined significance (MGUS) and from paired newly diagnosed and posttreatment multiple myeloma. The most significant difference between MGUS and newly diagnosed multiple myeloma (NDMM) samples was not related to cell density but to spatial heterogeneity, with reduced spatial proximity of BLIMP1<sup>+</sup> tumor cells to CD8<sup>+</sup> cells in MGUS compared with NDMM samples. Following treatment of patients with multiple myeloma, there was a reduction in the density of BLIMP1<sup>+</sup> tumor cells, effector CD8<sup>+</sup> T cells, and regulatory T cells, indicative of an altered immune microenvironment. Finally, bone heterogeneity decreased following treatment of patients with multiple myeloma. In summary, deep learning–based spatial mapping of bone marrow trephine biopsies can provide insights into the cellular topography of the myeloma marrow microenvironment and complement aspirate-based techniques.</p>Significance:<p>Spatial analysis of bone marrow trephine biopsies using histology, deep learning, and tailored algorithms reveals the bone marrow architectural heterogeneity and evolution during myeloma progression and treatment.</p></div>
<div>Abstract<p>Bone marrow trephine biopsy is crucial for the diagnosis of multiple myeloma. However, the complexity of bone marrow cellular, morphologic, and spatial architecture preserved in trephine samples hinders comprehensive evaluation. To dissect the diverse cellular communities and mosaic tissue habitats, we developed a superpixel-inspired deep learning method (MoSaicNet) that adapts to complex tissue architectures and a cell imbalance aware deep learning pipeline (AwareNet) to enable accurate detection and classification of rare cell types in multiplex immunohistochemistry images. MoSaicNet and AwareNet achieved an AUC of >0.98 for tissue and cellular classification on separate test datasets. Application of MoSaicNet and AwareNet enabled investigation of bone heterogeneity and thickness as well as spatial histology analysis of bone marrow trephine samples from monoclonal gammopathies of undetermined significance (MGUS) and from paired newly diagnosed and posttreatment multiple myeloma. The most significant difference between MGUS and newly diagnosed multiple myeloma (NDMM) samples was not related to cell density but to spatial heterogeneity, with reduced spatial proximity of BLIMP1<sup>+</sup> tumor cells to CD8<sup>+</sup> cells in MGUS compared with NDMM samples. Following treatment of patients with multiple myeloma, there was a reduction in the density of BLIMP1<sup>+</sup> tumor cells, effector CD8<sup>+</sup> T cells, and regulatory T cells, indicative of an altered immune microenvironment. Finally, bone heterogeneity decreased following treatment of patients with multiple myeloma. In summary, deep learning–based spatial mapping of bone marrow trephine biopsies can provide insights into the cellular topography of the myeloma marrow microenvironment and complement aspirate-based techniques.</p>Significance:<p>Spatial analysis of bone marrow trephine biopsies using histology, deep learning, and tailored algorithms reveals the bone marrow architectural heterogeneity and evolution during myeloma progression and treatment.</p></div>
Abstract Background Multiple Myeloma (MM) is a plasma cell malignancy that develops in the bone marrow. Function of T lymphocytes is impaired in patients with MM and the bone marrow microenvironment is described as hostile for T cell activity. Precise suppressive mechanisms within the bone marrow microenvironment remain poorly defined but will impact efficacy of bispecific T cell engager and chimeric antigen receptor (CAR) T cell therapies. Methods In this study T cell phenotype, function and metabolic activity were analysed within paired bone marrow aspirate and peripheral blood samples from 72 patients across the spectrum of MM, including individuals with premalignant and asymptomatic disease, alongside age-matched controls. This permitted assessment of effects of disease stage and the bone marrow microenvironment. The bone marrow microenvironment was also modelled in vitro using autologous plasma co-culture systems. Results Bone marrow CD8 + T cell function decreased with MM development and was consistently lower within bone marrow samples than matched peripheral blood. These changes were accompanied by decreased mitochondrial mass, which correlated tightly with T cell function. Conversely, long-chain fatty acid uptake and peroxidation was markedly elevated in bone marrow CD8 + T cells. In vitro modelling confirmed uptake of bone marrow lipids suppresses CD8 + T function, which was impaired in autologous bone marrow plasma, but rescued by both lipid removal and inhibition of lipid peroxidation. Analysis of single-cell RNA-sequencing data identified expression of fatty acid transport protein 1 (FATP1) in bone marrow CD8 + T cells in MM, and FATP1 blockade also rescued CD8 + T cell function. Finally, analysis of samples from treated patient cohorts identified CD8 + T cell metabolic dysfunction resolves in treatment-responsive but not relapsed MM patients and is associated with substantial functional restoration. Conclusions CD8 + T cells are functionally impaired within the MM bone marrow microenvironment. This is accompanied by decreased mitochondrial mass but elevated uptake of long-chain fatty acids. Blockade of FATP1 restores CD8 + T cell function in presence of BM lipids and may therefore represent a novel therapeutic target to augment their activity in the bone marrow in MM and improve efficacy of T cell-directed therapies.
Abstract Precision medicine holds great promise to improve outcomes in cancer, including haematological malignancies. However, there are few biomarkers that influence choice of chemotherapy in clinical practice. In particular, multiple myeloma requires an individualized approach as there exist several active therapies, but little agreement on how and when they should be used and combined. We have previously shown that a transcriptomic signature can identify specific bortezomib- and lenalidomide-sensitivity. However, gene expression signatures are challenging to implement clinically. We reasoned that signatures based on the presence or absence of gene mutations would be more tractable in the clinical setting, though examples of such signatures are rare. We performed whole exome sequencing as part of the CARDAMON trial, which employed carfilzomib-based therapy. We applied advanced machine learning approaches to discover mutational patterns predictive of treatment outcome. The resulting model accurately predicted progression-free survival (PFS) both in CARDAMON patients and in an external validation set of patients from the CoMMpass study who had received carfilzomib. The signature was specific for carfilzomib therapy and was strongly driven by genes on chromosome 1p36. Importantly, patients predicted to be carfilzomib-sensitive had a longer PFS when treated with carfilzomib/lenalidomide/dexamethasone than with bortezomib/carfilzomib/dexamethasone. However, in those predicted to be carfilzomib-insensitive, the latter therapy may have been capable of eradicating carfilzomib-resistant clones. We propose that the signature can be used to make rational therapeutic decisions and could be incorporated into future clinical trials.
<p>Performance evaluation of MoSaicNet and AwareNet deep learning models: <b>A</b>, The ROC curves and AUC values of the MoSaicNet superpixel classifier. The values in brackets indicate the 95% CI. <b>B,</b> Two-dimensional mapping of superpixels using MoSaicNet learned 200-dimensional features after dimensionality reduction by UMAP. <b>C,</b> The ROC curves and AUC values of single-cell classifier model on separately held test data. The values in brackets indicate the 95% CI. <b>D,</b> UMAP features visualization of deep learned features by AwareNet single-cell classifier CNN. <b>E</b> and <b>F,</b> Validation of AwareNet model using correlation of density of CD8<sup>+</sup> (<b>E</b>) and CD4<sup>+</sup> (<b>F</b>) cells in panel 1 and panel 2.</p>
<p>Computational methods for bone thickness analysis and cell infiltration patterns: <b>A,</b> Image analysis to estimate bone thickness (Supplementary Materials and Methods). Using the same BM sample image as <a href="#fig1" target="_blank">Fig. 1A</a>, the bone segmentation (ii) is an output of MoSaicNet (Supplementary Materials and Methods), and each bone is displayed in a different color. The color bar shows the pixel intensity of the image in iii and iv. The pixel intensity on the skeleton indicates half of the bone thickness (Supplementary Materials and Methods). <b>B,</b> Cell infiltration pattern analysis using NND and the null hypothesis of CSR (Supplementary Materials and Methods). Z < −1.96, Z > 1.96, and −1.96 ≤ Z ≤ 1.96 indicate a clustered, dispersed, and random distribution of observed cells, respectively. std, standard deviation; μ, mean NND of CSR.</p>
'%3e%3cpath fill='%23012169' d='M0 0v30h60V0z'/%3e%3cpath stroke='%23fff' stroke-width='6' d='m0 0 60 30m0-30L0 30'/%3e%3cpath stroke='%23C8102E' stroke-width='4' d='m0 0 60 30m0-30L0 30' clip-path='url(%23b)'/%3e%3cpath stroke='%23fff' stroke-width='10' d='M30 0v30M0 15h60'/%3e%3cpath stroke='%23C8102E' stroke-width='6' d='M30 0v30M0 15h60'/%3e%3c/g%3e%3c/svg%3e)
'%3e%3ccircle r='20' fill='%23008'/%3e%3ccircle r='17.5' fill='%23fff'/%3e%3ccircle r='3.5' fill='%23008'/%3e%3cg id='d'%3e%3cg id='c'%3e%3cg id='b'%3e%3cg id='a' fill='%23008'%3e%3ccircle r='.875' transform='rotate(7.5 -8.75 133.5)'/%3e%3cpath d='M0 17.5.6 7 0 2l-.6 5L0 17.5z'/%3e%3c/g%3e%3cuse xlink:href='%23a' transform='rotate(15)'/%3e%3c/g%3e%3cuse xlink:href='%23b' transform='rotate(30)'/%3e%3c/g%3e%3cuse xlink:href='%23c' transform='rotate(60)'/%3e%3c/g%3e%3cuse xlink:href='%23d' transform='rotate(120)'/%3e%3cuse xlink:href='%23d' transform='rotate(-120)'/%3e%3c/g%3e%3c/svg%3e)
