T-ALL leukemia stem cell 'stemness' is epigenetically controlled by the master regulator SPI1

Please note some of the gene names, such as PU.1, TIM-3, CD11c, have been changed to approved gene names systematically per editorial suggestions.

[…] While overall the findings are intriguing, the reviewers have suggestions to correct gaps in the experimental evidence provided. If the authors can address the issues noted below this could be a very interesting article.

1) Enthusiasm for the manuscript is limited by the lack of a compelling case for the relevance of the mechanisms proposed in human disease. KE-37 and CEM cell lines are analyzed as examples of PU.1⁺ and PU.1^- leukemia to illustrate the regulatory role of PU.1 on TIM-3 and CD11c, yet it is unclear that PU.1 expression or knockdown modifies the LIC activity in these lines or whether these lines really still have true LIC. Attention should be paid to the presence of MYC translocations and PTEN mutations in these lines. Correlative data on PU.1 TIM-3 and LMO2 in primary T-ALL samples is of interest but as the authors correctly and insightfully point out this correlation may reflect primarily developmental programs, which are potentially but not necessarily linked with LIC activity.

We appreciate reviewers’ critical comments on the human T-ALL cell lines used in our study. As indicated in the table below, KE-37 is PTEN null, NOTCH1WT and carries TRAD-MYC translocation, very similar to the genetic alterations associated with the Pten null T-ALL model used in our study, while CEM is PTEN null, NOTCH mutated and WT for c-MYC (Milani et al., 2018; Schubbert et al., 2014). We used these two lines to validate (1) the transcription regulation of the SPI1 target genes in human T-ALL (revised Figure 4I); and (2) the epigenetic regulation of the human SPI1 expression (revised Figure 10). Our recent analysis also indicates that the expression levels of SPI1 and HAVCR2 are higher in human ETP T-ALL as compared to non-ETP T-ALL (Author response image 1A). We concur that cell lines are most suitable for studying signaling and regulatory mechanism but not LIC activity.

Author response table 1.

Evaluation of the LIC activity in primary human T-ALL xenografts (PDX models) following the TIM-3 CD11c markers and ideally testing the effects of PU.1 inhibition in human T-ALL LIC activity would greatly enhance the relevance and general interest of the results presented here. We ask this in the hope that the authors may already have data in human T-ALL models, because this type of data would take more than a few months to generate, which is a goal of eLife. If such data is not available, the authors should at least point out the need for such studies and acknowledge the limitations of established T-ALL cell lines for studying the biology of LSC populations.

We totally agree with reviewer’s comment and have indeed generated six T-ALL PDX models. Unfortunately, none of PDXs growing up in the NSG mice so far belong to ETP T-ALL subtype, which our model closely mimics to (revised Figures 7 and 8). We have now collected more bone marrow samples and transplanted equal number of ETP and non-ETP T-ALL cells into NSG mice. Our preliminary study reveals that non-ETP cells may have better reconstitution ability than ETP T-ALL cells in NSG mice (Author response image 1B). Since ETP T-ALL is originated from early marrow or thymic progenitor cells before T cell commitment and its development depends on the functional thymic environment, NSG mice may not be suitable for studying human ETP T-ALL LICs without certain supplements or pre-requisitions, such as PTEN loss, which enables T cell differentiation in the absence of the recombination-activating gene (Reg) and substitutes for both IL-7 and pre-TCR signals (Guo et al., 2011; Hagenbeek et al., 2004).

Author response image 1

Download asset Open asset

There are very few publications on T-ALL LIC and T-ALL PDX models. The majority of these published works are based on non-ETP subtype and their conclusions about LIC are not consistent with each other (Chiu et al., 2010; Cox et al., 2007). We are conducting more detailed LIC-enrichment studies as well as genomic analyses of all T-ALL samples we have in hand. In the meantime, we are also exploring in vitro culture systems.

Since these works will take more than several months to accomplish, we have revised our manuscript and acknowledged the limitations of established human T-ALL cell lines for studying the biology of LICs (Subsection “LSCs loses their “stemness” when Spi1 expression is silenced by DNA methylation”, second paragraph and Discussion, last paragraph).

2) In Figure 3D, the authors use PU.1-EGFP for PU.1 ChIP analysis. Was it not possible to do the ChIP-PU.1 experiment without introducing PU.1-EGFP? Isn't there an antibody suitable to detect PU.1 in ChIP-seq? The authors should do a straightforward ChIP-seq experiment for endogenous PU.1 and detect the genes that bind PU.1 in the TIM-3^high mouse leukemia cell population. ChIP-PCR is subject to bias and it would be nice to show which of the 710 possible PU.1 target genes are actually bound by PU.1?

The frequency of HAVCR2^high subpopulation is about 1/10000 while regular ChIP-seq requires 10^6-7 cells. Since we can only sort about 4000 cells/hour, it is technically difficult to do the ChIP-seq analysis on endogenously isolated HAVCR2^high subpopulation. Our attempts of expanding HAVCR2^high subpopulation in vitro without changing its LSC activity have not been successful.

We also like to clarify the path that leads to the discovery of SPI1 in our study, which was not clearly described in our initial submission. We first used network component analysis, in which the activity of transcription factors can be deduced based on the expression levels of their target genes. Among the predicted transcription factors that may control the expression of HAVCR2^high LSC signature genes, SPI1 scores the highest. We then compared 1079 HAVCR2^high LSC-signature genes to 7180 T cell development related SPI1 target genes determined by the endogenous SPI1 ChIP-seq analysis (Zhang et al., 2012) and found 710 overlapping genes (please see revised description in the first paragraph of the subsection “SPI1 is the master regulator of LSC signature genes”).

3) Figure 3H. Please include Western blot for TIM-3.

We have now included Western blot for HAVCR2 in our revised Figure 4H.

4) Figure 5C. β-catenin is also known to upregulate MYC and yet the TIM-3^highPU.1^high LSC T-ALL subpopulation appears to have low levels of MYC expression due to PU.1 activation. Could the authors include Myc expression levels in Figure 5C? Does the Myc expression level go up or down when β-catenin is becomes activated by TIM-3?

As shown in the revised Figure 8A (original Figure 5C), Myc expression is indeed downregulated upon β-catenin activation in this in vitro system. MYC protein levels are also negatively correlated with the levels of non-phosphorylated β-catenin and HAVCR2 in vivo(revised Figure 2B and Figure 8C).

5) The authors propose a feed forward loop between PU.1/TIM-3 that is further regulated by active β-catenin, but the key data presented (Figure 5) are correlative and do not formally demonstrate that this is either a feed forward loop (i.e. that TIM-1 regulates PU.1 as well as the converse; nor that these are directly regulated by β-catenin). This needs to be strengthened or the concept of a feed forward loop should be downplayed in the interpretation.

We now have additional data to support the role of HAVCR2 in SPI1 regulation:

1) Treating the Pten null T-ALL model with the anti-HAVCR2 antibody for 2 days is sufficient to reduce HAVCR2^high subpopulation, accompanied by reduced non-phosphorylated β-catenin levels in vivo, similar to the effects of tankyrase/β-catenin inhibitor BAY6060 shown in revised Figure 8G and Figure 12A.

2) Gal-9 is a ligand of HAVCR2. Previous studies indicate that Gal-9 is critical for HAVCR2 signaling and human AML LSC self-renewal (Kikushige et al., 2015). We have genetically deleted Gal-9 and found that Gal-9 loss prevents the Pten null T-ALL development and HAVCR2^high LSC formation in vivo, an effect similar to that of Spi1 deletion (Author response image 2).

Author response image 2

Download asset Open asset

Nevertheless, we do agree with reviewer’s critical comment since β-catenin-SPI1-HAVCR2 forms an inter-dependent regulatory circuit and lacking anyone of them will influence LSC generation and self-renewal, as indicated by our Pten;Spi1 and Pten;Gal-9 knockouts and combination treatment experiments. In our revised Figure 13, we replaced feed forward arrows with a shaded triangle to more accurately present our take-home message. We also replaced “feedback loop” with “regulatory circuit” in our revised manuscript.

Stylistic comments:

1) eLife does not limit the number of figures. Also each figure can have linked supplemental figures. In the current paper, each of the figures has too many panels, such that the panels are very small and the font size is much too small. These panels should be broken up, with an increased number of main figures and figure supplements so that the results are more accessible. Please label the graphs with bigger font size – it is hard to read the labels.

We have revised figures per your instruction and the revised manuscript now contains 13 figures, 1 table and 4 supplementary figures:

– Revised Figure 1 is the original Figure 1.

– Revised Figures 2-3 and Table 1 are from the original Figure 2.

– Revised Figures 4 – 5 are from the original Figure 3.

– Revised Figure 6 is the original Figure 4.

– Revised Figures 7-8 are from the original Figure 5 and supplementary Figure 4.

– Revised Figures 9–10 are from the original Figure 6 and supplementary Figure 5.

– Revised Figures 11–13 are from the original Figure 7 and supplementary Figure 6.

2) Figure 2E is a table in a figure, and lists a lot of sensitivity, resistance data etc. that are not described.

The original Figure 2E has been moved to Table 1 with a clear description.

3) Overall the English in this paper is very understandable, but grammatical mistakes are found throughout the text. A scientific editor familiar with English grammar could very quickly improve the English grammar and clarity of the paper.

Thanks for pointing out this deficiency. The revised manuscript has been edited by Nature Research Editing Service.

References:

Hagenbeek, T.J., Naspetti, M., Malergue, F., Garçon, F., Nunès, J.A., Cleutjens, K.B.J.M., Trapman, J., Krimpenfort, P., Spits, H. (2004) The Loss of PTEN Allows TCR αβ Lineage Thymocytes to Bypass IL-7 and Pre-TCR–mediated Signaling. J Exp Med. Oct 4; 200(7): 883–894, doi: 10.1084/jem.20040495