Challenging the adult T-cell leukemia/lymphoma (ATL) stem cell concept based on limiting dilution transplantation assay

To the editor:

In their article, Yamazaki et al¹  reported the identification of malignant stem cells in the Tax-transgenic (Tax-Tg) mouse model of adult T-cell leukemia/lymphoma (ATL). This statement was grounded partly on a limiting dilution transplantation assay (LDTA) reported in Table 2 of their paper,¹  aimed at estimating the frequency of cancer stem cells (CSCs) among splenic lymphomatous cells (SLCs). From this assay, the authors reported that one CSC existed in 10⁴ SLCs (0.01%, ie, 1 in 10 000), and this frequency estimate was found to be consistent with the frequency of the CD38⁻/CD71⁻/CD117⁺ cells estimated by flow cytometry (0.03%, ie, 1 in 3333). Based on these cell frequencies, the authors claimed that CD38⁻/CD71⁻/CD117⁺ cells can stand for the rare stem cell population among the whole SLC population, and this hypothesis was documented by the ability of sorted CD38⁻/CD71⁻/CD117⁺ cells to regenerate the original lymphoma in transplanted animals in a single dose assay.

The rarity of cancer-initiating cells is usually considered as a major attribute favoring the cancer stem cell hypothesis.²  This perception of cancer-initiating cells' rarity, and the related stem cell concept, mainly relate to the estimation of cancer-initiating cell frequencies by the “gold standard” limiting dilution transplantation assay in recipient animals.³  However, the hypothesis of rarity of cancer-initiating cells is deeply impeded by the observation that LDTA practitioners, including Yamazaki et al,¹  usually do not use proper statistical analysis of LDTA data that may validate the reported cell frequency estimates. We are capable to demonstrate that appropriate statistical modeling of the LDTA performed with SLCs¹  invalidate this experiment and the related “stem” cell frequency estimate provided by the authors. Mathematical modeling of limiting dilution experiments of cells injected into recipient animals traditionally refers to the well-known single-hit Poisson model (SHPM), which posits that a single cell is necessary and sufficient to form a detectable tissue in the host.³  In a previously published paper⁴  we advised a statistical test aimed at estimating the fit of the SHPM to the data and based on a generalized linear modeling approach.⁵  Briefly, a general linear model, termed GLM_loglog, can be written as

Y_i = α + βX_i

with Y_i = ln [−ln (π_i)] and X_i = ln (x_i), where π_i is the proportion of recipient mice free of tumor, and x_i is the cell dose, that is, the number of cells injected into each mouse at each group i of cell dose. On elementary rearrangement⁴  of the above equation, the GLM_loglog reduces to the SHPM for the special case as the slope β = 1. Therefore, testing the null hypothesis β = 1 explores the SHPM hypothesis, and this can performed by a standard likelihood ratio test^5,6  and by a standard Wald test.⁴  The results of our modeling study are presented in Table 1. The null hypothesis β = 1 is clearly rejected at a very significant level, indicating that the LDTA does not conform to the SHPM, and the consequence is that no frequency estimate can be rendered. The precise frequency of tumorigenic cells in the Tax-transgenic (Tax-Tg) model remains unknown, and the hypothesis that this frequency may be considerably higher than 0.01% must not be ruled out, ultimately challenging the stem cell concept.