Quantified Morphology in Diagnosis of Hematologic Malignancies

In 1845, Rudolf Virchow described and named the disease of leukemia using light microscopy. He also emphasized that diseases including cancers, originated from normal cells.¹  Since then, huge advances in cell-based cancer detection approaches revolutionized diagnosis and prevention of human cancers. Imaging methods, next-generation sequencing of cancer cells, and identification of circulating tumor cells from blood biopsies provide an opportunity to predict and monitor tumor progression and therapeutic responses.^2-7  However, cancer detection is still a major issue for pathologists and physicians, and in many cancers, light microscopy and immunophenotyping are still the gold standards of the initial diagnosis, especially for pediatric lymphomas and leukemias.

Microscopy and flow cytometry have been extensively used for diagnosis of hematologic malignancies, but these modalities have limitations that make them insufficient for a precise diagnostic classification. Microscopy is a skill-based and subjective approach that is considerably variable between pathologists, even those with years of training. In contrast to microscopy, flow cytometry can detect different cell types according to the expression of cluster of differentiation (CD) molecules on the cell surface. Although flow cytometry is a powerful method in detection of various cell types in normal samples, it cannot distinguish cells with similar or overlapping immunophenotypes, such as tumor cells that display similar surface markers to their normal counterparts.

Knowing the unique capacities and limitations of morphology and immunophenotyping, Dr. Albert G. Tsai and colleagues developed a high-throughput and multiplexed morphometric assay that not only identifies different blood cell lineages in normal and malignant bone marrow samples but could distinguish tumor from normal cells using subcellular features. Investigators used antibody-measurable cellular antigens to quantify common cell morphological features such as chromatin, cytoplasm, and vesicles. To quantify the intracellular structures in hematopoiesis and hematologic malignancies, authors selected 11 common structural components of cell morphology corresponding to chromatin quality, nuclear shape, nucleolar size, granularity, granular color, cytoplasmic color, and cell size; they tested these in the bone marrow biopsies of healthy donors and in a group of 71 diverse clinical samples including acute myeloid leukemia, acute lymphoblastic leukemia, myelodysplastic syndromes, myeloproliferative neoplasm, B- and T-cell lymphoma, and multiple myeloma. Authors used CD markers to detect various cell types in a normal bone marrow, and then identified the morphometric signals in each. Although cellular structures are ubiquitously found in different cell types, morphometric analysis of bone marrow populations showed numerous differences between various lineages. Nuclear protein lamin-B1 and ribosomal 5.8s rRNA were highly common in blast (progenitor) cells. Cytoskeletal β-actin was enriched in neutrophils and monocytes, and heterochromatin protein 1β was found to be associated with neutrophil differentiation. Lymphocytes and erythroids were morphometrically similar; thus, authors used a CD45 marker to distinguish these populations. Being able to morphometrically define major cell types, Dr. Tsai and collogues analyzed the single-cell morphometric features of malignant samples and classified different cell populations with almost similar morphometry to normal samples. Although combination of morphometric markers could not distinguish normal from neoplastic cells, individual measurement of morphometric signatures revealed significant differences in normal and malignant bone marrow samples. Nuclear membrane proteins lamin A and C distinguished normal from neoplastic mature T cells, and a combination of lamin B1 and ribosomal RNA differentiated normal and leukemic blast cells independent of flow cytometric markers such as CD34 and CD117. They also identified the VAMP7-A morphometric marker as a substitution for light-based side scatter in diagnostic cytometry.

Combining multiple single-cell morphologic signatures with a machine learning approach (supervised dimensionality reduction by linear discriminant analysis [LDA]), authors generated a morphometric map (MM) that recapitulated and improved upon traditional light-scatter gating strategies. The morphometric LDA method was also able to visualize continuous and branching processes such as gradual differentiation, which is not detectable in diagnostic cytometry. Additionally, MM could distinguish tumor blasts from normal blasts in myeloid leukemias with much higher efficiency compared to flow cytometry and light microscopy, suggesting a clinically practical method that is easy to interpret and could be extended to other datasets.