News & Views | Published:

Cancer

A gene-expression profile for leukaemia

Nature volume 540, pages 346348 (15 December 2016) | Download Citation

Can simple genetic risk profiles be identified for complex diseases? The development of a gene-expression profile for acute myeloid leukaemia suggests that they can, and that they may improve prognosis prediction. See Letter p.433

On page 433, Ng et al.1 report a tool that improves the prediction of prognoses for people who have a form of acute leukaemia. The researchers began by identifying populations of cells that exhibit key properties — collectively known as stemness — that enable the cells to initiate and sustain leukaemia. This allowed the authors to ascertain gene-expression profiles for stemness, and to use them as the basis of a scoring system for risk. The work demonstrates how gene-expression profiles can be used to enable reliable prognoses for complex diseases.

Acute myeloid leukaemia (AML) is characterized by the presence of a huge range of chromosomal and molecular aberrations. This means that there are many subgroups of people with AML who have widely different prognoses2. These groups are known as risk groups, and are used to determine which consolidation treatment should be given following initial chemotherapy (induction therapy). For example, transplantation of stem cells from donors is an option for patients judged to be at the most risk from the disease. But such transplantation often has fatal side effects3, so is not the best choice for some patients. Improvements to risk assessments are necessary, not only to make decisions about consolidation strategies, but also to choose between different types of induction therapy (which are expected to become available in the future).

Gene-expression profiles could be instrumental in realizing these improvements4. Ng and colleagues' approach, which relies on identifying profiles for stemness5, is a good example of how such profiles could be used. In normal tissue, stemness allows stem cells to self-renew — to sustain the long-term process of normal cell differentiation. In the haematopoietic system, normal haematopoietic stem cells (HSCs) are the origin of differentiated blood and bone-marrow cells. HSCs express CD34 proteins on their surfaces, but not CD38 proteins, and are thus said to have the CD34+CD38 immunophenotype. Leukaemic stem cells (LSCs) have stemness properties similar to those of HSCs, but they can express different patterns of cell-surface proteins: they can be CD34+CD38 cells (which are probably derived from HSCs), but they can also have CD34+CD38+, CD34CD38+ or CD34CD38 immunophenotypes. It has previously been shown in animal models that the leukaemia-initiating ability of these different CD34/CD38 subpopulations can differ6.

In a huge effort, Ng et al. isolated 227 CD34/CD38-defined cell fractions from 78 people with AML, and injected the fractions into mice (Fig. 1). They confirmed that the leukaemia-initiating ability of the cell fractions differed: leukaemia could form from all the cell fractions obtained from a patient, from some of the fractions or from none. The authors then compared gene expression in the original cell fractions that caused leukaemia with gene expression in cell fractions that did not, irrespective of the cells' CD34/CD38 immunophenotype. This allowed them to identify gene-expression patterns that were directly related to the ability of cells to form leukaemias in vivo in mice.

Figure 1: A 17-gene score for assessing the risk of acute leukaemia.
Figure 1

Ng et al.1 took cell samples from people with acute myeloid leukaemia (AML) and divided them into fractions based on the expression of CD34 and CD38 proteins on the cells' surfaces. The researchers transplanted the fractions into mice, and identified which fractions caused leukaemia and which did not. The authors then compared gene-expression patterns in the disease-causing cell fractions with those in the non-disease-causing fractions, and thus identified candidate genes that correlated with tumour formation. This information was used to direct a statistical analysis of gene-expression data that had previously been gathered in a clinical study7 of people with AML. The analysis identified a score that could be calculated for patients based on the expression of 17 genes. The score provides a reliable system for assessing patients' prognosis.

Ng and colleagues first identified 104 genes for which expression levels differed by at least twofold in leukaemia-initiating cell fractions compared with fractions that didn't initiate leukaemia. The authors then examined a large set of gene-expression data obtained from a clinical study7 of 495 people with AML, and found that 89 of the 104 genes were present in the set. The cells in that study were not divided into fractions, but displayed gene-expression patterns that were similar to those observed by Ng et al. in leukaemia-initiating cell fractions.

Next, the authors used a statistical method to relate gene expression to clinical outcome for these 89 genes, and for a subset of 43 genes that are highly expressed in leukaemia-initiating cell fractions. This allowed them to identify an optimal panel of 17 genes, the expression of which was highly indicative of a poor clinical outcome in a patient subgroup. The authors confirmed this finding in other AML cohorts and found that a scoring system based on their gene panel (called LSC17) offered superior prognoses when compared with other gene-expression profiling systems for AML5,8. In fact, Ng and colleagues found that previously reported genetic signatures of AML were not independent prognostic factors when tested in the other cohorts.

Ng et al. also found that gene-expression patterns associated with stemness in AML are independent of the chromosomal and molecular aberrations used to assess patient risk, showing that stemness is a factor that crosses the borders of previously identified risk groups. Finally, the authors developed an assay that allows gene-expression data to be rapidly generated, which could form the basis of a fast (24–48 hours) prognostic test for patients.

As the authors indicate, analysis of large data sets from clinical studies in which both extensive information about the mutational status of leukaemia cells9 and LSC17 scores are available will be needed to assess whether the prognostic value of the LSC17 score is independent of the prognostic value of mutations present at diagnosis. The clinical benefits of the LSC17 score must be assessed, because prognostic value does not always lead to a meaningful clinical advantage. Moreover, small populations of leukaemia cells that have a similar genetic make-up (clones) can be present at diagnosis, survive therapy and proliferate to cause a relapse (in some cases after having acquired additional mutations10,11). Only time will tell whether Ng and co-workers' gene-expression profiles account for cell fractions defined by such clones, and thereby predict associated relapses.

The prognosis of a person with leukaemia at the point of diagnosis is only part of the prognostic story. Once treatment has started, factors such as therapy compliance, alterations to drug doses that are made to mitigate side effects, and differences between patients in the concentration of drugs in blood plasma might partly override the effects of prognostic diagnosis parameters such as gene-expression patterns. The consideration of post-treatment parameters such as measurable (minimal) residual disease12(a measure of the persistence of small numbers of leukaemia cells in patients in remission) has drastically changed the landscape of risk assessment in AML. Assessing combinations of cellular properties at diagnosis, non-cellular patient-specific factors during therapy, frequencies and properties of cells that remain after treatment and changes in immunological parameters, might offer a more-refined prognosis than is currently possible. This would enable more-personalized induction and consolidation treatments to be used. Ng and colleagues' study is potentially a big step towards such assessments, especially at the diagnosis stage.

Notes

References

  1. 1.

    et al. Nature 540, 433–437 (2016).

  2. 2.

    et al. Blood 116, 354–365 (2010).

  3. 3.

    et al. Nature Rev. Clin. Oncol. 9, 579–590 (2012).

  4. 4.

    & Exp. Hematol. 42, 651–660 (2014).

  5. 5.

    et al. Cell 162, 184–197 (2015).

  6. 6.

    et al. J. Clin. Invest. 121, 384–395 (2011).

  7. 7.

    et al. Haematologica 94, 131–134 (2009).

  8. 8.

    et al. J. Am. Med. Assoc. 304, 2706–2715 (2010).

  9. 9.

    et al. N. Engl. J. Med. 374, 2209–2221 (2016).

  10. 10.

    et al. Nature 481, 506–510 (2012).

  11. 11.

    & Oncogene 32, 135–140 (2013).

  12. 12.

    et al. Semin. Hematol. 52, 184–192 (2015).

Download references

Author information

Affiliations

  1. Gerrit J. Schuurhuis is in the Department of Hematology, Free University Medical Center, De Boelelaan 1117, 1081HV Amsterdam, the Netherlands.

    • Gerrit J. Schuurhuis

Authors

  1. Search for Gerrit J. Schuurhuis in:

Corresponding author

Correspondence to Gerrit J. Schuurhuis.

About this article

Publication history

Published

DOI

https://doi.org/10.1038/nature20488

Further reading

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.

Newsletter Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing