Methodologies and Emerging Technologies for the Evaluation of the Hematopoietic System

Abstract

Hematology and bone marrow analysis is central to our understanding of the hematopoietic system and how it responds to insults, and this session presented during the 2022 STP symposium provided a review of current and novel approaches for the evaluation of the hematopoietic system in the context of nonclinical investigations. This publication summarizes the information presented on novel approaches for evaluation of the hematopoietic system using automated hematology analyzers, including details around the quantitative assessment of bone marrow cell suspensions as well as introducing several newly available hematology parameters. It was followed by a discussion on intravital microscopy and live cell imaging and how these methods can assist with de-risking hematopoiesis-associated safety concerns, and a review of recent assays using artificial intelligence for the evaluation of bone marrow.

Novel Methods for the Evaluation of Hematopoiesis and Circulating Blood Cells

Traditional methods for the evaluation of hematopoietic cells consist of peripheral blood cell counts, histological and cytological evaluations of hematopoietic organs. Hematology analyzers offer a large spectrum of core hematology parameters that are gaining interest, but a large number of measurements are not transferred from the hematology analyzers to the Laboratory Informatic System, because only well-known and understood parameters are reported. In this presentation, we have chosen to discuss reticulocyte maturation parameters available with the Advia®2120i- Siemens or Sysmex XN-1000-V™ analyzers, and the automated quantitative evaluation of bone marrow cell suspensions on the Sysmex XN-1000-V™ analyzer, and to review the methods and indications of such evaluations.

For the evaluation of reticulocyte maturation, there are different parameters or indices available on the Advia®2120i- Siemens or Sysmex XN-1000-V™ analyzers. The mean corpuscular volume (MCV) is indicative of the size of all red blood cells independently of their stage of maturation, but the MCVr refers only to the size of reticulocytes, and the MCVm to the size of mature red cells. Similarly, the mean hemoglobin concentration (MCH) represents the content of hemoglobin of all stages of red cells including reticulocytes, but the reticulocyte hemoglobin concentration (CHr) is measured on the Advia®2120i- Siemens and similarly on the Sysmex XN-1000-V™ analyzer (Ret-He). The mean corpuscular hemoglobin concentration (MCHC) can also be separated for mature red cells (CHCMm) and reticulocytes (CHCMr) on the Advia®2120i- Siemens analyzer. Reticulocyte maturation indices are measured on both analyzers: immature reticulocytes have a high content of RNA that decreases with maturation: the different fractions High (HRF), Medium (MRF), and Low (LRF), are reported by both analyzers, and the Immature Reticulocyte Fraction (IFR) representing the sum of the High and Medium fractions, is reported in the Sysmex XN-1000-V™ analyzer. All these indices are part of routine hematology analysis, although only a few of them are reported in nonclinical studies.

Reticulocyte Hemoglobin Content as an Early Predictive Biomarker of Iron Deficiency

An important value of a biomarker is its ability to predict risk of a disease prior the disease onset. Decreases in Ret-He∕CHr correlate with early decreases in iron availability, independently of the cause of iron deficiency. There are now several published case reports observing very early Ret-He∕CHr decreases, several weeks before the occurrence of iron deficiency anemia which becomes evident only in the later stages of iron deficiency. Hypochromic microcytic reticulocytes due to reduced hemoglobin synthesis are observed before changes of mature red blood cells and before alteration of iron serum levels. In short term studies evaluating iron deficiency anemia (2 weeks), it was possible to observe decreases in Ret-He∕CHr along with MCVr and CHr, while MCVm remained unchanged. ¹ Dose-responsive decreases in Ret-He∕CHr were reported to occur simultaneously with decreased brain iron content in rat pups. ² In longer-term studies evaluating iron deficiency anemia (4 weeks and more), decreases in RBC, MVC, MCH, and serum iron were observed.

Automated quantitative evaluation of bone marrow cell suspensions with the Sysmex analyzer is another new method of interest for the hematology laboratory. The Sysmex XN-1000-V™ differentiates cells through light-scatter similarly to flow cytometry, and by RNA and DNA fluorescence staining. It offers the unique possibility to gate targeted cell populations for the analysis of particular body fluids, and it can be customized for automated quantitative evaluation of bone marrow cell suspensions, although the qualification of the method is necessary and should be qualified in each laboratory. Bone marrow gating is species-specific, and a new qualification is necessary for any new species evaluated, but once established, the gating of bone marrow is stable and needs no further adjustment. It can also be used across studies. This method has initially been qualified with success and published for Wistar rats ³ and then for Monkeys, dogs and mice. ⁴ We were able to reproduce a similar qualification for Sprague Dawley (SD) rats.

The principles remain similar: initially a bone marrow suspension is first analyzed to determine the Total Nucleated Cell Count (TNCC). For SD rats, a femur is collected at necropsy and flushed with 5 mL of fetal bovine serum. The sample is washed at least once and homogenized by repeated aspirations using a 1-mL syringe and a needle, until the bone marrow aggregates are dissociated. The sample is diluted in phosphate buffered saline (PBS; 50 mL), and the total number of cells counted (TNCC) in the bone marrow suspension is determined on the Sysmex analyzer. With that method, the TNCC ranged between 1000 and 3000 cells/µL for healthy SD rats. A cytocentrifuge slide was prepared for each sample for evaluation. The analyzer provides a cytogram, but the different cell boxes are not assigned to any bone marrow cell population. In order to identify the different types of cells on the cytogram, we used anti-CD11b and anti-biotin rat magnetic microbeads to separate the myeloid cells and to determine the cloud of the myeloid precursors on the cytogram. Similarly, anti-CD45RA (lymphoid cells) and anti-CD3 (lymphocytes T) rat magnetic microbeads were used to capture lymphoid cells (mature lymphocytes and lymphoblasts) and determine their localization on the cytogram. Lymphocytes and mature erythroid cells are very similar in size and density, and cannot be differentiated by the instrument, and the separation of the cells will be necessary for all bone marrow samples evaluated on the instrument. Once the erythroid and myeloid cell gatings are defined, the gating separation between proliferating and maturating precursor cells is set up for myeloid and erythroid cells by comparison with microscopic 300-cell bone marrow differentials.

There are benefits to pursuing quantitative evaluation of the bone marrow by Sysmex. The bone marrow differentials obtained by Sysmex in rats were similar to those obtained by microscopic observation, and since slide quality was not a factor, the cell differentials were consistent and independent of operator. The method improves timeliness of cytologic examination as it limits the microscopic evaluation to the observation of morphological changes and megakaryocytes. The TNCC could also potentially be used as an assessment of bone marrow cellularity when bone marrow collection is well standardized.

Although quantitative evaluation of the bone marrow by Sysmex can provide benefit in some circumstances, it is a complex, multistep process, and a resource-intensive addition to standard nonclinical studies. Therefore, it should be used only when it can add value. Importantly, this evaluation cannot be added after hematotoxicity is observed in a study; it must be built into the study protocol beforehand. Analysis may be triggered following a prior study in which changes in hematology were unexplained or insufficiently characterized by bone marrow histopathology. Situations in which this type of advanced marrow evaluation might be considered include enhanced characterization of altered marrow cellularity, examination of early hematopoietic precursors, investigation of atypical cells, and differentiation between erythroid and lymphoid cell populations. In addition, quantitative bone marrow assessment may also be added on a case-by-case basis, depending upon knowledge about the therapeutic target, for instance.

Applying Artificial Intelligence to Bone Marrow Evaluation

Xenobiotic compounds, including those for immunotherapy and chemotherapy, can affect hematopoietic precursor cell populations resulting in cellularity changes in bone marrow. Anatomic pathologists evaluate bone marrow using hematoxylin and eosin (H&E) stained sections of bone and assign subjective severity scores to findings. Clinical pathologists evaluate bone marrow using bone marrow smears at high magnification to identify precursor cells and determine the ratio of myeloid to erythroid cells (M:E ratio). Artificial intelligence (AI)-based tools can be developed to identify, quantify, and differentiate cell types in the bone marrow which can assist clinical and anatomic pathologists as screening tools and decision aides.

We developed AI models using the Aiforia and the Deciphex Patholytix platforms. In both platforms, developing a model involves a training process and a validation process. The training process is accomplished by using annotations that teach the AI a learning objective, based on an established “Ground Truth,” and forms the basis of the subsequent output of the model. For example, in order to teach the model to identify hematopoietic cells, specific annotations are created by the trainer to “show” the model which cells are hematopoietic cells and which are not, in a manner similar in spirit to training a pathology resident. The platform’s software develops the model once annotations are complete. The model may be further refined through additional annotations if necessary. A model with good performance can proceed to the validation process. Validation compares the performance of AI models to human pathologists. The training and validation processes are similar between the Aiforia and Patholytix platforms.

The Aiforia model was trained to identify and count the total number of hematopoietic cells in whole slide images (WSIs) of cynomolgus macaque sternebra. ⁵ Aiforia uses “layers,” in which the initial layer is trained to identify a tissue. Next, subsequent layers are trained, the second layer identifies components of the tissue in the initial layer and the third layer identifies subcomponents in second layer, and so on. We trained 3 layers. The first was the total bone to delineate the entire sternebra. We then trained a second layer to identify the marrow compartment (sternebra minus bone). Next a third layer was trained to identify hematopoietic cells (marrow compartment minus adipocytes and blood vessels). Finally, an object counter was used in the third layer to identify and count individual hematopoietic cells. The validation process for Aiforia is a feature of the software. Areas for validation, known as Validation Regions of Interest (ROIs), were defined using the Aiforia software tool. Both the AI model and human pathologist evaluate the ROIs, in this case counting the cells in the ROI, and the results are compared. In our model, the validation results determined the AI model’s performance to be similar to that of human pathologists. The model was successful in counting hematopoietic cells and provided the total surface area of the bone layer, from which a cell density (cells/mm²) could be calculated. We compared the cell densities calculated from the AI model to severity scores from a study pathologist in a study that demonstrated decreased hematopoietic cellularity in the bone marrow. We found the cell densities inversely correlated to the severity scores (increased cell density correlated to lower severity scores), suggesting this model could be a useful screening tool in bone marrow evaluation. ⁵ We further trained the model to distinguish myeloid and erythroid precursor cells. The myeloid-erythroid-ratio (M:E) could be calculated, however, the AI-driven histologic M:E ratio did not correlate to the manual cytologic M:E ratio performed on bone marrow smears from the same animal. The usefulness of the AI-driven histologic M:E ratio requires further evaluation.

Efforts are ongoing using the Patholytix platform for bone marrow evaluation. Patholytix uses annotations to train the model for different “classes.” For bone marrow we used 5 classes. The classes were “tissue” (bone, cartilage, and blood vessels), erythroid cells, myeloid cells (including lymphocytes), megakaryocytes, and adipocytes. There was also a “background” class that identified all the space on the slide that was not tissue. Like the Aiforia platform, the model was taught using a ground truth established by trained pathologists to provide examples of tissue belonging to each class. The final output of the model was a colored overlay, called a “inference mask,” in which different colors represent the classes or tissue/cell types. Pathologists can easily screen bone marrow slides in Patholytix by selecting and viewing all bone marrow slides and assessing the relative amounts of each color in the masks. The transparency of the mask can be altered to evaluate and verify the results of the mask, potentially drawing attention to abnormal findings. Early results show the Patholytix AI bone marrow model demonstrated acceptable performance with minimal confusion between erythroid and myeloid cells.

Both the Aiforia and Patholytix AI models provided useful tools for screening bone marrow for potential lesions and assisting in decision making. Both were developed with limited resources in time and personnel. These models lay the foundation for more advanced models in the future. In addition, data generated from AI models, such as cell density of bone marrow, which have not been available before, have value that is yet to be determined.

Intravital Microscopy and Live Cell Imaging—Neutrophil Extracellular Traps and Cellular Toxicity

In the past decade, there have been several exciting developments in live cell imaging and automated microscopy that have greatly enhanced our understanding of cellular processes. Although a plethora of novel methods have been used to elucidate cellular signaling pathways, identify and decipher biological processes, much of the knowledge has been derived from these methods is static which does not necessarily reflect the dynamics of cellular processes. It is in this context that intravital imaging with multiphoton microscopy provides helpful insights into the dynamics of cellular processes that cannot be studied by other techniques.

Intravital microscopy (IVM) refers to imaging of live animals or cells that allows observing biological processes at microscopic resolution. IVM is also described as intravital imaging or live cell imaging. IVM often involves 2-photon microscopy, in which simultaneous excitation of 2 photons with longer wavelength than emitted light allows imaging of live tissues up to hundreds of microns in thickness ⁶ and high content or high-resolution imaging which applies automated fluorescent microscopy, fluorescent detection, and multiparameter algorithms to visualize cellular processes.

Unlike traditional histology which involves fixation, mechanical sectioning and staining of tissues, IVM uses fluorescent labeling and optical sectioning capabilities of multiphoton and confocal microscopes in intact, live tissues which enables observing cellular dynamics and cell–cell interactions visualized in real time. ⁷ Furthermore, 3-dimensional reconstruction of the tissue is feasible with IVM by computationally combining imaging data from multiple stacks of images.

The technical details on considerations for IVM, principles involved and detailed applications of IVM is beyond the scope of this section. Several well-referenced reviews provide additional information on the principles, applications of IVM^7-11 and the application of IVM in drug discovery research.^12,13 Regarding its research applications into hematopoietic system, IVM has been a very useful tool in providing insights neutrophil extracellular traps (NET) and the involvement of platelets in this phenomenon.

Crucial to their role in innate immune system, neutrophils release large, extracellular web-like structures named as NETs composed of cytosolic and granule proteins that trap, neutralize and kill bacteria, fungi, and viruses.^14,15 Platelets are critical in the production of NETs through their ability to bind to neutrophils and induce NET release.^16-18 Activated platelets and/or complement can induce NET formation and NETs, in turn can lead to activation of complement and coagulation cascades. ¹⁹ Activated platelets interact with neutrophils through exposed surface P-selectin or through secreted high mobility group box protein 1 (HMGB1), which in turn activates neutrophils and induces NET formation. ²⁰ NETs can activate coagulation pathways either through expression of tissue factor, activation of FXII or degradation of TFPI. ²¹ NETs complexed with platelets can thus, form a scaffold for thrombus formation and complement activation. This interaction between platelets, neutrophils, complement and coagulation systems is instrumental in the pathogenesis of several inflammatory diseases.

In summary, IVM is a powerful technique to nondestructively visualize cellular and subcellular structures. Combining this with macroscopic imaging can lead to powerful correlative imaging that has potential to improve mechanistic insights from single cell to a whole organism level.