Autologous hematopoietic stem cell transplantation (HSCT) is a therapy routinely used to treat malignant blood diseases and is extended to other diseases. It is a complex, specialized, high-cost treatment with a high degree of personalization. From 1957 to 2016, approximately 1,298,897 hematopoietic transplantation (57.1% autologous) procedures were recorded worldwide.1 Unfortunately, approximately a third of autologous or allogeneic transplant recipients die for reasons other than the relapse of the primary disease, including engraftment or organ failure, perhaps due to biases in the collected subpopulations.2

The number of CD34+cells/kg of weight to be infused is one of the most critical factors that define the success of this type of therapy.1 However, the phenotypic composition of the hematopoietic stem cell (HSC) population must be clearly defined. Since the 1980s, the procedures for HSC mobilization and collection techniques in peripheral blood have made it possible to reduce the complexity of the process that was performed using bone marrow.3 At the same time, the development of single-cell characterization techniques shows more heterogeneity in this population, making the definition of the HSC slightly inconsistent.4

In addition to the CD34+cells/kg dose, some subsets have been added as predictive biomarkers: CD34+/CD38−, CD34+/CD90− and CD34+/CD38+.5,6 However, it is necessary to increase the number of markers in the immunophenotyping since it allows for deepening the diversity of the HSCs, as they are a heterogeneous group of cells. Subsets characterized by two markers do not consider rare or minority populations that play an essential role in the homeostasis of bone marrow cell niches or ignore their parental phenotypes. For this reason, it is necessary to extend immunophenotyping, taking advantage of technological development and the flexibility of this methodology. Finding regularities within this diversity will allow for the discovery of new and more robust biomarkers. This emerging knowledge makes the treatment approach more complex, but it allows for the identification of possible biases within the balance of progenitors of stem lineages collected to define the prognosis with greater certainty and provide more elements for the follow-up of these patients. However, the “gold standard” for HSCT remains the quantification of progenitor cells with the CD45dim/CD34+ phenotype and characterization of the apheresis product yield, primarily using the International Society for Hematotherapy and Graft Engineering guidelines (ISHAGE).7,8 Nevertheless, it results in a reductionist vision of the quality assurance of the product to be transplanted. The clinical implications of knowing the sub-phenotypic composition of autologous HSCs in the apheresis product are evident.

Biases within these subpopulations of hematopoietic progenitors may lead to the engraftment failure of specific lineages.9 In this sense, identifying imbalances in the composition would make it possible to recover fractions by modifying or repeating the mobilization and collection scheme until the optimal concentrations of each affected subpopulation are reached, before performing the transplant. Studies on multiplexed platforms will be required to define predictive biomarkers and to know optimal concentrations.

Mass cytometry (MC) is one of the methodologies that has allowed multiparametric analysis on a single cell. More than 40 antigens can be identified in the same cell without some of the complexities of flow cytometry (FC), such as compensation and fluorescence-minus-one controls. However, it has limitations, such as not offering an absolute count of cells/µL and not having the complete standardization and validation process for clinical diagnosis.10 Multiparameter analysis strategies should be completed with unsupervised machine learning analyses, such as the FlowSOM, that allow for the elimination of bias introduced by studies based only on known phenotypes.11 As the number of markers on a single cell increases considerably, automation is needed for a more suitable approach.

Current evidence suggests that it should not be enough to quantify the collected stem cells. They should be characterized to detect lineage biases and, possibly, functional alterations, senescent phenotypes and pathological populations contained in the product to be transplanted, as all of them will affect the success of the transplant. Therefore, this study aimed to assess the feasibility of using the MC to characterize and enumerate the CD45dim/CD34+ stem cell population collected by apheresis, from patients with chronic lymphoid malignancies undergoing autologous HSCT, at the Abu Dhabi Stem Cells Center (ADSCC), United Arab Emirates.

The increase and diversification of hematopoietic transplantation require a better understanding and characterization of the set of stem cells collected for transplantation to predict its success and prognosis. Based on the known heterogeneity of the HSC as a complex system,12,13 it is essential to identify lineage biases in the product to be transplanted. In this sense, the methods of choice are the multiparametric and multidimensional analyses, allowing for the advantage of the MC benefits and potential.14,15 However, this method has yet to be approved for use in the clinical practice of hematopoietic transplantation. This study provided scientific evidence that suggests the feasibility of starting the validation process of this technology, so further studies that validate its use in the clinical setting are required. Previous reports have demonstrated its usefulness in evaluating the immune profile after hematopoietic transplantation and graft-versus-host disease, supported by the high-dimensionality approach to identify potential predictor biomarkers.16–18

In the present study, a correspondence study was initially carried out between the gold standard (FC) and the MC to approach the use of the MC in the highly-concentrated HSC sample quantitative immunophenotyping, as occurs in apheresis collections from mobilized peripheral blood. Our results show that it is possible to use the MC to quantify the HSC, as the correspondence and association analyses between the FC and MC have shown outstanding agreement, even when it was required to use the absolute count by the double platform in the case of the MC. This correspondence was demonstrated by the Bland-Altman method. In addition, statistical significance tests were used to validate the result, which showed a higher strength of the correlation (r = 0.98/94 and R2 = 0.86/87; relative/absolute counts, respectively). The slight decrease in the correspondence found for the absolute counts could be due to using a double platform in the MC, differences in sample staining protocols and cell loss, that occur with MC washes.

Previous studies also report excellent correlations between both methodologies. Nicholas et al.,19 reported perfect correspondence between both methods for immunophenotyping common and rare populations in PBMCs. Recently Ravkov et al.,20 conducted a study of many samples with excellent results. In their article, the authors state that a crucial element was to design an optimal window strategy that would allow the analysis to be performed similarly. We chose a system that followed the logic of the first proposal of these authors, that of using the same sequence order. Still, multiple internal controls were added to eliminate unwanted events and the viability control required by ISHAGE was maintained using cisplatin. The MC has the advantage that two iridium isotopes (191Ir and 193Ir) are also included in the panel design for double labelling of the DNA, which makes it possible to detect nucleated events with precision and to discriminate doublets.21 Another element that distinguishes the present study is the inclusion of the absolute count by a double platform of the HSC in the apheresis bag, with an excellent correspondence with the FC.

After checking the possibility of quantifying the CD45dim/CD34+ target population, immunophenotyping of its subsets was performed with their relative and absolute concentrations. The concentration of cells per microliter is essential to calculate the dose in the apheresis bag, which is expressed in cells x 106/kg. In this sense, the referential values ​​are reported. However, no studies have been performed to establish doses by subpopulations. Our global values ​​of CD45dim/CD34+coincide with those reported in previous studies.22–24

Even after mobilization, the low frequency of the HSC subpopulation in peripheral blood renders their immunophenotyping challenging. However, apheresis bags are more feasible, as they are more concentrated, because apheresis works as an enrichment method. Their analysis is better and more reliable, even applying the MC.25 Nevertheless, there is no consensus regarding the CD34+ pool subpopulation immunophenotypes. Recent models consider the complexity of the hematopoietic system and they propose a continuous, flexible and dynamic landscape in which many variables mediate in the ontogeny of stem cells derived from bone marrow.12,26–28 We suggest seven immunophenotypes characterized by nine markers described in the literature (Table 1) that were analyzed in the Kaluza.27,29–41

Although without significant differences, the studied subpopulation distribution using the manual gating in the apheresis bags samples showed a higher percentage and absolute count of the late CMP and MLP progenitors over the HSC and MPP early ones, which could be related to the action of the mobilizing agent G-CSF. (Figure 3A) The subset most committed to the lymphoid lineage (PreB/NK) was the lower represented, perhaps due to the same action of the G-CSF mobilizing agent. Previous studies have shown that G-CSF can affect the composition of HSC subpopulations, causing an imbalance towards less primitive progenitors.42,43 However, in our study, no significant imbalance could be due to the effect of plerixafor, which tends to do the opposite, favoring early phenotypes due to the changes it induces in bone marrow niches.43

The HRC subpopulation showed higher concentrations with statistically significant differences because they were defined as a non-specific HSC (CXCR4+) phenotype that was calculated over total CD34+ cells. The median percentage value of the HRC population was similar to that which was previously described by Dlubek et al.,44 who report that in mobilized blood and apheresis this phenotype represents almost half of the total CD34+ cells. However, our study range was extensive (2.1% - 60.4%), possibly due to the different mobilization regimens applied, in which plerixafor was included in almost 30% of the patients. Pharmacological studies of this drug showed that it blocks the binding of the clone (12G5)45,46 used in our study to detect the CXCR4. However, in the cases that only used filgrastim, the median values for this subpopulation (data not shown) were like those previously reported.

Another population detected with the markers used was the aPC, found in low concentrations in only four patients in our study. Contamination of the apheresis product in multiple myelomas with this rare population of plasma cells was reported many years ago.38,47 Higher apheresis product purity has been associated with more profound responses, whereas contamination decreases overall survival and progression-free survival, in multiple myeloma. However, some studies have found a higher probability that recurrence occurs from residual cells in the marrow, thus weakening the role of the aPC contamination in engraftment.48-50 Although they report their existence and collection for transplantation, the reports of Kuranda et al.51 on multiple myeloma patients state that more studies are required to verify their potential adverse effect on post-transplanted patients.

The aPC is recognized as an independent prognostic factor and may be a new additional tool to demonstrate the efficacy of induction regimens. Therefore, its quantification is recommended, not only in bone marrow, but also in apheresis products.48-50 Quantifying the aPC in the graft also allows patients with high levels to receive additional cycles of induction treatment, reducing graft contamination and overall tumor burden prior to the autologous HCT.49,50

The identification of rare populations, such as the aPC, is favored by unsupervised multiparametric cytometric analyses.

The HSC lineage biases could occur during the emergency hematopoiesis after transplantation and should be considered, as it can lead to transplant failure due to multiple causes.12,52 One of the most critical moments of identifying and treating lineage biases occurs during the characterization of the apheresis product, as complementary apheresis can be indicated to reach the optimal dose for a specific subpopulation. In this regard, there is evidence that the global dose collected critically influences the composition and lineage bias of the HSC population.53 Therefore, it is another element to consider for the subpopulation quantification and the dose calculation before concluding the number of aphereses planned for each patient. Another element in favor of the search for lineage biases is the underlying disease, which affects the bone marrow and hematopoiesis in the case of myeloma, and lymphoma may affect it.54 The bone marrow infiltration can generate these lineage biases. Once identified, although it cannot be corrected in this case, the probability of failure could be evaluated according to its magnitude. In addition, the possibility of the allogeneic variant of transplantation could be assessed and a chimerism obtained would solve the bias, without needing a complete allogeneic transplant. This latter should first be evaluated in laboratory animal models. As for the distribution of progenitors according to lineage commitment in our samples, no significant differences were found among the subpopulations, suggesting they were balanced and there were no lineage biases. These results showed a trend comparable to the reports by Elsayed et al.55. Although the subpopulations and phenotypes are not precisely the same from a hierarchical consensus point of view, they are the same.

The approach to the complexity and heterogeneity of the stem cell population was complemented with an unsupervised machine learning analysis to eliminate possible manual processing biases. Fewer groups of subpopulations were defined by these methods than by the classical manual. Still, the main phenotypes were conserved, as seen in the heat map of the unsupervised analyses of automatic and manual gating. (Figures 4C and C, respectively) The reduction in subpopulations could be due to the expression pattern of some secondary markers and the overlap among subpopulations of HSC described in current models.12,26,28 As previously mentioned, the mobilization process could have contributed to the expression modification of the antigens and their phenotypes.43 These possible causes are described in the literature. They may cause a decrease in the definition of marker expression patterns, which is why automated analyses are recommended to eliminate selection bias.

The dimensionality reduction images offered an excellent visualization of the expression and distribution of the markers. (Figures 4A, 4B, 5A, and 5B) The separation of some subpopulations and the overlap of others showing heterogeneous groups was evidenced. Similar results have been found with the same unsupervised analysis strategy, but with different technologies, such as transcriptomics using single-cell RNA sequencing expression profiling.56

Among the study's limitations are the small sample size and the reduced number of markers used according to the MC potentiality. In this same sense, the Human Hematopoietic Stem and Progenitor Cell Phenotyping Expansion Panel Kit were required to be completed with two markers, CD38 and CD45, to increase the specificity of the proposed phenotype. The line of research will continue to increase the number of markers to phenotype cells, take advantage of the potential of the MC in the approach to high-dimensionality complex analysis and evaluate the clinical implications of future findings.

Another limitation that must be faced is the cost of this technology. However, the first step to reducing the cost and generalizing its use is to demonstrate its usefulness by emphasizing its advantages. There is a precedent for this problem: flow cytometry presented the same challenge in its beginnings. One of the strategies could be to locate this technology initially only in reference centers where samples are sent for analysis. This study offers clues to the benefits of the MC and its combination with machine learning (ML) methods that justify the cost-benefit balance. Among these advantages are:56,57

• •

  Handling large volumes of data: Multiparameter flow cytometry can generate huge amounts of data. In combination with the ML, the MC makes it possible to process, analyze and extract valuable information from these large amounts of data, which would be extremely difficult or even impossible to do using conventional methods.

• •

  Identification of complex patterns: Subtle and complex patterns are detected in the data, even those that may be missed by manual analysis. This can reveal intricate relationships between different markers and cells, enriching our understanding of cell biology and improving the classification of cell subpopulations.

• •

  Reduced human bias and improved reproducibility: Manual analysis may be subject to inter- and intra-observer variations and biases. The ML provides a more objective and consistent approach to data analysis, improving the reproducibility and validity of the results.

• •

  Automation of repetitive tasks: the ML can automate routine tasks in multiparameter cytometry analysis, freeing researchers from the manual workload and allowing them to focus on more creative and strategic tasks.

• •

  Antibody Panel Optimization: The ML can help design efficient, customized antibody panels to study specific cell populations or traits, maximizing the experiment throughput and reducing costs.

• •

  Time and resource efficiency: The automation provided by the ML in the multiparameter cytometry analysis saves time and resources, allowing researchers to focus on more creative and complex aspects of the research.

In conclusion, the MC is useful in studying subpopulations of the HSC collected in apheresis products. It is feasible to accurately enumerate CD34+ cells and their subpopulations using a dual platform, as it has shown excellent agreement with the gold standard and allowed for the identification of rare subsets. The use of unsupervised algorithms satisfactorily completes and validates the multidimensional analysis. The strategy can be extended to apheresis products from conditions other than the AHSCT.

Carlos Agustin Villegas-Valverde: Conceptualization, Methodology, Supervision, Data curation, Formal analysis, Writing – original draft, Writing – review & editing. Antonio Alfonso Bencomo-Hernandez: Conceptualization, Methodology, Supervision, Data curation, Formal analysis, Writing – original draft, Writing – review & editing. Yandy M. Castillo-Aleman: Methodology, Supervision, Data curation, Formal analysis, Writing – review & editing. Yendry Ventura-Carmenate: Conceptualization, Data curation, Formal analysis, Writing – review & editing. Imilla Casado-Hernandez: Data curation, Formal analysis, Writing – review & editing. Rene Antonio Rivero-Jimenez: Data curation, Formal analysis, Writing – review & editing.