We read with interest the paper of Glimm and Eaves,1 in which the effect of in vitro division on the function of primitive human hematopoietic progenitor cells (HPC) was examined. The interest stemmed from the fact that this paper investigated an issue that we previously examined in a series of 5 papers2-6 and that was also addressed by other laboratories in a similar fashion.7,8 Data from Glimm and Eaves1 confirmed results we first described in 1995,2 which were similar in scope to previous9 and subsequent findings7,8illustrating that a small fraction of cultured human CD34+ cells remains quiescent in culture with associated cytokine nonresponsive (CNR) characteristics while maintaining primitive hematopoietic potential. Although the paper by Glimm and Eaves reiterates and corroborates previously published results from many laboratories, the authors construed their results to support specific conclusions that may not be directly borne by the experimental evidence presented in their report.1 This letter is intended to demonstrate that, viewed differently, data presented by Glimm and Eaves can be interpreted to be in full support of the previous work rather than in disagreement with earlier findings that were directly and strongly criticized by the authors.
At the center of this controversy is the hematopoietic potential of groups of cells of different proliferative history following their maintenance in ex vivo expansion cultures. Specifically, these are cells that we previously defined as cytokine nonresponsive (CNR) cells,2 which remain undivided for several days in expansion cultures, as compared with those that undergo 1, 2, 3, or more cell divisions, which Glimm and Eaves referred to as postmitotic.1 We described in several publications2,4,5 that the frequency of long-term hematopoietic culture-initiating cells (LTHC-ICs)2 was higher among CNR cells than among those proliferating extensively in expansion cultures. Similarly, Glimm and Eaves demonstrated that, in at least 3 out of 6 samples examined, the frequency of long-term culture-initiating cells (LTC-ICs) was several-fold higher in the undivided fraction than in the postmitotic fraction.1(Table 2) When cells that had undergone no more than 2 divisions were tested against those dividing at least 3 times, the frequency of LTC-ICs among the former group was in excess of 80-fold higher in 4 of 5 samples tested.1(Table 3) In their discussion, Glimm and Eaves argue that in our studies,2 in which a similar approach was taken to isolate cells of different proliferative history, we may have isolated, in the same group, cells undergoing up to 4 divisions (rather than CNR cells only) due to the “lower resolution of PKH2 staining.”1(p2165) According to Glimm and Eaves, the insensitivity of our protocol may explain why an increased frequency of LTHC-ICs was detected among cells we “thought to have remained quiescent”1(p2165) in culture.
It is unfortunate that Glimm and Eaves did not examine all of the published literature on this same issue. In a paper published in 1998,6 we clearly demonstrated that the frequency of LTHC-ICs, although relatively high among CNR cells (0 divisions), was highest among cells undergoing 1 division in vitro. LTHC-IC frequency among cells dividing twice was as high as that observed for nondividing cells (CNR cells). Only when cells divided more than 3 times did the frequency of LTHC-ICs decline considerably. That we were able to dissect the proliferative history of cultured cells into single cell division cycles and to examine the frequency of LTHC-ICs among each fraction demonstrating the persistence of LTHC-IC frequency through 2 divisions confirms 3 important points: first, that our methodology was sensitive enough to examine single divisions rather than groups of cells undergoing no more than 2 or at least 3 divisions as achieved by Glimm and Eaves1; second, that in light of our previously published results,6 where cells were grouped based on single rather than multiple divisions, results reported by Glimm and Eaves are more limited and less informative; and third, and most importantly, that our data demonstrated loss of hematopoietic activity through the first few divisions, as reconfirmed by Glimm and Eaves, making therefore their attempt to interpret these results as an endorsement for maintenance of function through “multiple”1(pp 2161-2162) self-renewal divisions unjustified.
In the second part of their paper, Glimm and Eaves transplanted fractions of cultured cells to estimate the number and distribution of NOD/SCID competitive repopulating units (CRUs) among progeny cells. However, test cells were competed against 106 irradiated human cells, thus negating the “competitive”1(p2162)nature of the assay and turning it into a simple assessment of engraftment potential of groups of cultured cells. Although Glimm and Eaves may argue that endogenous murine stem cells may act as competitor cells, such an assumption has not been formally proven to impact the outcome of xenogeneic engraftment in this model. It is therefore important when the results of these transplants are examined, as in Glimm and Eaves's Table 5, to carefully consider the end point of such an assay. Under these conditions, the comparison shifts from being a true assessment of the frequency of CRUs to a comparison of the engraftment ability of 1.8-7.8 × 103 cells for cells that underwent no more than 2 divisions and in excess of 8.1 × 105 cells for cells that underwent at least 3 divisions. Although the difference in the size of the graft between the 2 groups was on average 207-fold in favor of the at-least-3-divisions group, a substantial fraction (7 of 29; 24%) of transplanted mice in the no-more-than-2-divisions group displayed multilineage engraftment at levels comparable to those seen in 14 of 28 (50%) mice receiving cells that underwent at least 3 divisions. Once again, Glimm and Eaves interpreted these data to indicate that “multiple” self-renewal divisions had taken place among cells that underwent at least 3 divisions.1(pp 2161-2162) In a murine transplantation model,3 we previously examined the same issue. In a setting where it was possible to evaluate and compare numbers of transplanted cells, we demonstrated that the relative frequency of long-term repopulating cells was higher among CNR cells than among postmitotic cells, although both groups of cells were compromised in their repopulating potential relative to fresh cells.3 Transplantation studies of Glimm and Eaves1 yielded the same results, namely, that the frequency of CRUs in cells that divided less than 3 times was 1 per 14 000 cells, while that among cells that had divided at least 3 times was 1 per 1 300 000 cells. Thus, in spite of similarities in both the design and results between our previous study3 and that of Glimm and Eaves,1 data from the latter group were construed differently by the authors.
It was also interesting to closely examine some of the values used by the authors to calculate the frequency of CRUs among cells that underwent no more than 2 or at least 3 divisions. As indicated by Glimm and Eaves,1(p2165) data from “all like experiments” were used for limiting dilution analysis; so it is safe to assume that results appearing in their Table 5 were also used. In experiments 4, 5, and 6 of Table 5, a decrease in the frequency of engrafting animals (66.6% to 33.3% to 16.7%) was observed when an escalating number of cord blood cells from no more than 2 divisions were transplanted (from 6.2 to 6.6 to 7.8 × 103cells, respectively). Given that these results are the inverse of those expected in a limiting dilution analysis and that in other groups of test cells only 2 data points were available for analysis, conclusions drawn from these calculations are severely undermined. The implication of such a shortcoming is profound. Based on these calculations, the authors declare the “total calculated change in total CRU numbers in these experiments was an ≈ 2.4-fold increase.”1(p2165) Interestingly, it is not stated how the 2.4-fold increase was calculated: neither the frequency nor the total number of CRU detected before expansion of test cells is provided. If these calculations are not definitive, for the reasons discussed above, it is likely that the reported 2.4-fold increase may be overestimated. That being the case, and in the absence of any meaningful increase in the numbers of CRU, it is doubtful that any self-renewal divisions of human hematopoietic cells were achieved, let alone “multiple” divisions of this nature.
In summary, we believe that the data presented by Glimm and Eaves1 in fact provide further supporting evidence for a series of previous publications2-8 and confirm the general conclusions reported in those papers. We regret that Drs Glimm and Eaves interpreted their results differently and concluded that their findings were in disagreement with previous data. Nevertheless, it is refreshing to witness continued interest and new studies in an area of investigation that we fostered for a number of years. Undoubtedly, these efforts should yield new and valuable information and help clarify many unresolved issues.
High resolution CFSE labeling demonstrates self-renewal divisions of human in vivo repopulating cells in short-term culture
We are pleased to hear of Edward Srour's interest in our recent paper.1-1 But we believe his letter misrepresents the purpose of the work reported in our paper and contains a number of other points of confusion that warrant clarification.
The first and probably most significant issue concerns important technical differences between our studies and those dealing with human cells referred to by Dr Srour.1,2 This includes potential differences in the cell populations identified as long-term culture-initiating cells (LTC-ICs) in our studies and as long-term hematopoietic culture-initiating cells (LTHC-ICs) in the studies of the Srour group, which are detected by an assessment of the colony-forming cell (CFC) content of LTCs maintained for different periods of time (ie, 6 weeks versus 2-5 weeks, respectively) and using different feeder and growth-factor conditions. We and others1-5-1-7 have previously shown that prolongation of LTC-IC assays (performed using stromal feeders with or without growth-factor provision) results in the detection of quantitatively and qualitatively different LTC-IC subpopulations, with a marked change occurring between 5 and 6 weeks. Even more important is the fact that the growth factors used in our short-term cultures included 100 ng/mL Flt3-ligand (FL) in addition to Steel factor (SF) and interleukin-3 (IL-3) (and interleukin-6 [IL-6] and G-CSF),1-1 whereas FL was not used as part of a cocktail in the comparable studies by the Srour group.1-2,1-4 Again, as we and others showed using LTC-IC amplification endpoints1-8,1-9 and single-cell1-10 or high-resolution carboxyfluorescein diacetate succinimidyl ester (CFSE)–tracking of CD34+CD38− cell proliferation,1-11 the additional presence of FL is critical to an enhanced rate of recruitment into division of the most primitive types of human hematopoietic cells, including those present in cord blood.1-12 This underlies the improved retrovirus-mediated gene transfer efficiency to stem cells obtained when FL is added to SF, IL-3, and IL-6 cocktails used to stimulate them,1-13,1-14because transduction using such vectors requires passage of the target cells through mitosis. Thus most CD34 cells, including many LTC-ICs assessed using a 5-week assay, can be mitogenically activated by SF and IL-3,1-15,1-16 but more primitive cells require the additional presence of FL to maximize their mitogenesis with preservation of their original functional activity.1-9 1-10
A second issue raised by Dr Srour concerns the precision of different methodologies for resolving populations that have executed 0, 1, 2, 3, or more divisions. This is key to the interpretation of studies where functionally defined cells are present at low frequencies in the populations being tracked. Because our major objective was to track the proliferative activity of cells with in vivo repopulating ability, we felt it was important to maximize the resolution of different generations of cells by preselection of a very narrow window of CFSE-labeled cells prior to placing them in culture, so that any potential overlap between cells assigned to a given CFSE-defined population would be minimal (see, for example, Figure 1 in Glimm and Eaves1-1). In the absence of any independent measure of cell division, we would continue to question the resolving power of distributions such as those shown in Figure 1 of Traycoff et al1-2 or in Figure 1 of Traycoff et al.1-4 The fact that these authors observed differences between cells categorized as having executed 0-5 divisions does not address the issue of their purity in this regard.
We would also point out that our decision to compare only 2 subsets of each cultured population was unrelated to the resolving power of the separation. Quite to the contrary, it was dictated by our primary intent to investigate the proliferative history of cells retaining in vivo repopulating activity and the practical limitations of such experiments when we did not know beforehand what the answers were likely to be.
A third issue concerns the interpretation of data obtained from our NOD/SCID repopulation measurements. First, it should be noted that the limiting dilution assay used has been shown to measure human competitive repopulating unit (CRU) frequencies independent of the number of non-CRU human cells injected,1-17 as long as sufficient human cells are included in low cell innocula to prevent nonspecific losses.1-18 We have used the term “competitive” to include situations where the competition is not defined (but constant and possibly zero). We agree that the competitive activity of the endogenous murine stem cells has yet to be characterized. But the fact that human cell engraftment decreases rapidly as a function of decreasing doses of radiation used to pretreat the recipients1-19 is certainly consistent with some competition being provided by the murine cells. Because in our studies more than 80% of all the cultured cells were used to determine the distribution of CRUs between the 2 fractions compared in each experiment, it is the total yield of CRU per fraction, not the CRU frequency, that is the important comparison. Such a comparison clearly shows that no CRUs remained in the small residual population of viable cord blood or fetal liver cells that had not divided after 5 days of exposure to FL, SF, IL-3, IL-6, and G-CSF and that a substantial proportion of the total CRUs detectable in such cultured populations had divided at least 3 (ie, multiple) times. We have used the term “self-renewal” on the assumption that cells detectable as CRUs after culture would have been derived exclusively from cells detectable as CRUs prior to culture (difficult to prove but, nevertheless, a likely assumption given that the cultures were initiated with FACS purified CD34+lin− cells).
Dr Srour also expresses concern with our use of limited data sets to derive CRU frequencies and the observation of apparently opposite trends in the proportions of engrafted mice in subgroups that were transplanted with slightly different numbers of cells in different experiments. In fact, the ranges in cell doses cited in our Table 5 (Glimm and Eaves1-1) vary by only 25%. Such a difference could not be discriminated unless extremely large groups of animals were compared. Indeed, we believe these data indicate a convincing reproducibility of the results obtained from independently executed experiments using similar numbers of cells obtained from different donors. We therefore felt well-justified in pooling the data from “like” experiments. This was confirmed by the methodology used to calculate CRU frequencies (which requires that the data show an adequate fit to a Poisson distribution) and the accuracy of the CRU frequencies derived, as defined by the corresponding SEM values.
The final issue concerns the general observation that CRU numbers are not increased as much as would be expected if all self-renewal divisions yielded 2 daughter CRUs, or even if the frequency of such events occurred more often than not. It is likely that multiple explanations underlie the failure to observe significant net increases of CRUs in vitro. Indeed, the studies of Gothot et al1-3indicating changes in CRU repopulating activity with their passage from G0 into G1 have made an important contribution in this regard. But it should also be remembered that extensive amplification of a small subset of LTC-ICs/CRUs can occur even when the total LTC-IC/CRU population shows a significant net decrease.1-20,1-21 Moreover, the criteria we have used to define CRUs do not allow qualitatively compromised subsets to be detected (if such are generated) when CRU amplification has been stimulated either in vivo1-22 or in vitro.1-17 1-23
In summary, we believe the enhanced resolving power of the CFSE-tracking method employed in our studies coupled with the use of statistically reliable limiting dilution measurements of human repopulating stem cells provides strong and novel evidence not only that such cells can divide several times within 5 days in culture but also that more than 90% of all such cells detectable in such cultures are the progeny of cells that have divided at least once. We believe the interesting question now is, therefore, not why they don't divide but why they fail to maintain transplantable stem cell activity more often than not.
