they undergo homeostatic proliferation3–5,8,9. The conversion of Treg cells to other types of effector T cell — if it occurs readily and fre­quently — could be harmful, because the resulting effector T cells, like the Treg cells from which they are derived, are likely to recognize self antigens, and so may cause autoimmune disease.

Immunologists are on the quest to assess this possible danger. There are, however, conflict­ing results from experiments in mice that are engineered to express a reporter dye in their T cells whenever the cells express Foxp3; the dye­expressing T cells are then traced to deter­mine whether they lose Foxp3 expression and become effector T cells. An earlier report6 showed that a significant fraction of Foxp3+ Treg cells (about 10%) stop expressing Foxp3 and differentiate into TH cells that secrete the pro­inflammatory cytokine IFN­γ and that mediate type 1 diabetes in a mouse model of autoimmune disease. Rubtsov et al.2, however, show that Treg cells are highly stable in terms of both Foxp3 expression and their suppressive function, with few cells converting into effec­tor T cells even after exposure of the animals to an inflammatory cytokine milieu or to X­ray radiation, which causes a reduction in the level of circulating T cells.

Although apparently conflicting, these findings provide insights into how T­cell dif­ferentiation to and from Treg cells depends on Foxp3 expression, on the mix of cytokines and on the presence of other T cells (Fig. 1). For one thing, a paucity of IL­2 clearly plays a cru­cial part in Treg conversion: when Foxp3+ Treg cells are transferred to T­cell­deficient mice, the co­transfer of T cells that do not express Foxp3 or infusion of IL­2 prevents conver­sion of the Foxp+ Treg cells to effector T cells. And if Treg cells are transferred on their own, Foxp3­negative T cells that have formed from Foxp3+ Treg cells produce IL­2 , which inhibits the conversion of further Foxp3+ Treg cells in a negative­feedback loop3. Similarly, in Rubtsov and colleagues’ study2, the sublethal X­ray radiation given might reduce T­cell levels sub­stantially but not completely, allowing residual T cells to secrete IL­2 and thereby inhibit Treg conversion.

Another possibility is that Foxp3+ T cells vary in functional stability and therefore in their susceptibility to conversion; that is, only a certain fraction of Foxp3+ Treg cells may be ‘plastic’. In support of this notion, the regula­tory regions of the Foxp3 gene are more exten­sively demethylated in thymus­derived Treg cells than in those induced by TGF­β10, offer­ing an explanation for the functional instability of the latter population and its higher suscep­tibility to conversion. Even thymus­derived Foxp3+ Treg cells could vary in the methylation status of their Foxp3 gene, in their suppressive activity and in their eventual fate. Indeed, there is evidence11 for functional and phenotypic variability in human Foxp3+ T cells: Foxp3high

cells are terminally differentiated to be highly suppressive, whereas some Foxp3low cells are non­suppressive and can secrete effector cytokines.

Finally, alterations in the level of Foxp3 expression may affect the functional stability of Treg cells and, possibly, their susceptibility to conversion. For example, mice that express a reporter dye along with Foxp3 in their Treg cells show a slight reduction in Foxp3 expression and an increased susceptibility to autoimmune disease12. This might explain the discrepancies in Treg­cell conversion between the two stud­ies2,6, which used different genetic manipula­tion methods such that different levels of the reporter dye were expressed.

Thus, the plasticity of T cells to differenti­ate to and from Foxp3+ Treg cells is elaborately controlled by factors internal and external to these cells. Collectively, these factors ensure a remarkably constant number of Foxp3+ Treg cells in the immune system (about 10% of all T cells expressing the surface marker CD4), with a general increase only occurring at sites of inflammation. Further understanding of the molecular basis of functional stability and the

plasticity of Treg and other T cells should facili­tate safe and effective control of physiological and disease­associated immune responses. ■

Shimon Sakaguchi is in the Department of Experimental Pathology, Institute for Frontier Medical Sciences, Kyoto University, Kyoto 606-8507, Japan, and is also at the WPI Immunology Frontier Research Center, Osaka University, Japan. e-mail: shimon@frontier.kyoto-u.ac.jp

1. Sakaguchi, S., Yamaguchi, T., Nomura, T. & Ono, M. Cell 133, 775–787 (2008).

2. Rubtsov, Y. P. et al. Science 329, 1667–1671 (2010).

3. Duarte, J. H. et al. Eur. J. Immunol. 39, 948–955 (2009).

4. Tsuji, M. et al. Science 323, 1488–1492 (2009).5. Oldenhove, G. et al. Immunity 31, 772–786 (2009).6. Zhou, X. et al. Nature Immunol. 10, 1000–1007

(2009).7. Bettelli, E., Korn, T., Oukka, M. & Kuchroo, V. K.

Nature 453, 1051–1057 (2008). 8. Yang, X. O. et al. Immunity 29, 44–56 (2008).9. Wan, Y. Y. & Flavell, R. A. Nature 445, 766–770

(2007). 10. Huehn, J., Polansky, J. K. & Hamann, A. Nature Rev.

Immunol. 9, 83–89 (2009).11. Miyara, M. et al. Immunity 30, 899–911 (2009).12. Wing, K. et al. Science 322, 271–275 (2008).

D i a B e T e S

Podocytes lose their footing Impaired insulin action, combined with its insufficient secretion, can cause diabetes. In a surprising extension of this notion, decreased insulin action in the kidney’s podocyte cells may contribute to renal complications in diabetes.

C H r i S T i a n r a S k- m a D S e n & G e o r G e l . k i n G

The hormone insulin is best known for regulating glucose and fat metabolism in skeletal muscle, fat tissue and the

liver. Impaired insulin action in these tis­sues is central to the development of type 2 diabetes, which can be debilitating or even fatal, owing to kidney failure or cardiovascu­lar complications. During the past decade or so, responses to insulin have been identified in other cell types, where some of its effects are completely different from those regulating metabolism. Writing in Cell Metabolism, Welsh et al.1 show that, in mice, loss of insulin action in podocytes — a specialized type of kidney cell — can cause changes in kidney structure and function that resemble the renal compli­cations of human diabetes known as diabetic nephropathy.

One of the kidney’s main functions is to regulate water and salt balance in the body. To this end, the kidney performs ultrafiltra­tion of blood plasma: every day, some 70 litres

of filtrate pass through the kidney’s collecting ducts, while plasma macromolecules such as albumin are retained. Plasma filtration takes place through loops of capillaries in micro­scopic structures called renal glomeruli. These capillaries are covered by podocytes, which are epithelial cells that coat the entire capillary surface with the foot­like processes that give them their name. The filtration barrier consists of a single layer of endothelial cells — the cell type that forms the inner lining of all blood vessels — separated from the podocytes by a basal membrane.

One of the earliest abnormalities in the development of diabetic nephropathy is short­ening of podocyte foot processes, which con­tributes to the breakdown of the glomerular filtration barrier, thereby allowing albumin and other macromolecules to escape into the urine. Albumin loss in the urine and other renal abnormalities, which in some cases progress to kidney failure, also frequently occur in patients with metabolic syndrome — a com­mon condition, associated with obesity and

4 2 | N A T U R E | V O L 4 6 8 | 4 N O V E m b E R 2 0 1 0

NEWS & VIEWSRESEaRch

© 20 Macmillan Publishers Limited. All rights reserved10

impaired insulin action. Drugs that decrease blood glucose, lower blood pressure or inhibit the actions of the hormone angiotensin can delay, but not eliminate, the onset of diabetic nephropathy.

To relay its signal from the bloodstream to a cell’s interior, insulin binds to receptors on the cell surface. Welsh et al.1 used mice in which the gene encoding the insulin receptor was deleted specifically from podocytes. At birth, kidney appearance in these animals was nor­mal. At five weeks of age, however, they began to show excretion of albumin in the urine, shortening of the podocyte foot pro cesses, increased deposition of components of the basal membrane, and a higher frequency of programmed podocyte death through apop­tosis. Some animals even developed shrunken kidneys with prevalent scar tissue similar in appearance to the kidneys of humans with late­stage diabetic nephropathy.

These findings, however, are not just notable for their striking similarity to the pathology of diabetic nephropathy in humans. Welsh and colleagues’ mice also showed mild worsen­ing of kidney function. This observation is intriguing because, in the most commonly studied rodent model of diabetes, destruction of insulin­producing cells with the drug strep­tozotocin causes no significant change in kid­ney function, despite resulting in microscopic

kidney abnormalities and albumin excretion in the urine2.

Welsh et al. also report that insulin can reorganize the actin cytoskeleton in podo­cytes maintained in culture. This phenom­enon resembles the actin remodelling seen when insulin causes translocation of glucose­transport proteins to the cell surface in fat or skeletal muscle cells3. Exactly how regulation of the cytoskeleton affects both podocyte foot processes and the filtration barrier requires more detailed investigation.

Remodelling of the actin cytoskeleton also does not explain the increased apoptosis of podocytes lacking insulin receptors. In this regard, the observation4 that insulin enhances the expression of the protein vascular endothe­lial growth factor (VEGF) — a crucial survival factor as well as a regulator of blood­vessel for­mation — might be of relevance. Podocytes are the main source of VEGF in the kidney: mice lacking VEGF specifically in podocytes show partial loss of all major cell types in the glomerulus, including podocytes5,6. Moreover, VEGF expression is reduced in tissues such as heart muscle in animals with diabetes7. Insulin can also prevent apoptosis by other mechanisms8,9, including inactivation of the transcription factor FoxO and inhibition of caspase­9, a signalling molecule that promotes apoptosis.

Whether insulin signalling to other cell types in the glomerulus is essential for main­tenance of the filtration barrier is not known. Dysfunction of endothelial cells in the sys­temic circulation is associated with the initia­tion and progress of diabetic nephropathy, and endothelial cells respond to insulin by chang­ing the production of factors that regulate blood­vessel tone and by decreasing oxidant production and increasing levels of antioxidant enzymes10.

It also remains to be seen whether podo­cytes, or other renal cells, are insulin resistant in diabetes and metabolic syndrome in other animal models and in humans. It could be that some cell types or insulin­signalling pathways are more susceptible to insulin resistance than others. For example, insulin increases sodium transport in the kidney’s tubular cells, but this aspect of its function is not affected in diabetes11. To understand whether insulin resistance in other renal cells contributes to diabetic nephropathy, researchers must study normal insulin action in kidney cells; whether these cells develop insulin resistance in meta­bolic syndrome and/or in diabetes; and what causes impaired insulin signalling.

With its focus on insulin resistance in glomerular cells, Welsh and co­workers’ paper1 helps to establish that diabetic neph­ropathy — the leading cause of chronic kidney

Structural biologists are good at producing static snapshots of proteins, but seeing them in action is the ultimate goal. This is exactly what Kodera et al. have achieved in a remarkable study that appears elsewhere in this issue (N. Kodera, D. Yamamoto, R. Ishikawa and T. Ando Nature 468, 72–76; 2010). They have been able to directly visualize myosin V, a cytoskeletal motor protein, as it ‘walks’ along actin filaments.

The authors have developed a high-speed atomic force microscopy (HS-AFM) method that enables them to generate rapid images of proteins at much higher resolution than light microscopy. Here they apply this technological advance to visualize the stepwise movement of myosin V. The protein consists of two heads connected by long necks (lever arms) to a globular tail domain through a coiled-coil helix. The tail binds to

various cargo molecules to transport them along the actin filaments, the energy required being generated by hydrolysis of ATP to ADP.

The HS-AFM images showing the leading (L) and trailing (T) heads of the tail-truncated motor can be seen in Figure 1a of the paper on page 73, part of which is reproduced here. In frame 1, myosin is advancing from the left; in frames 2 and 3, it has marched into full view, the two heads and necks now being visible; in frame 4, it has moved on a step. See go.nature.com/DHBOY9 for the full movie.

The authors’ analysis confirms the known behaviour of myosin motors, including the use of successive 36-nanometre steps along the actin filament and the ‘hand-over-hand’ movement of the motor heads. However, the study also provides convincing evidence for the hypothesized ‘lever-arm swing’. In this feature, tiny changes in

myosin's head domain are amplified by the lever arm to produce large displacements at the far end of the neck that translate into movement of the whole protein along the actin filaments.

The new analysis also uncovers novel characteristics of the myosin V mechanism, such as a ‘stomping’ behaviour, in which either the L or T head becomes detached then rebinds to actin. The stomp is observed more frequently for the L than for the T head, but the T-head stomp often leads to a forward movement on the actin filament.

The insight into the behaviour of myosin V that Kodera et al. reveal will have a major impact on understanding the mechanisms involved in biological molecular motors. More generally, the technological advance provided by HS-AFM looks set to take a prominent place in the field of biomolecular imaging. Deepa nath

C e l l B i o l o G Y

Myosin in motion1 2 3 4

TL

4 N O V E m b E R 2 0 1 0 | V O L 4 6 8 | N A T U R E | 4 3

NEWS & VIEWS RESEaRch

© 20 Macmillan Publishers Limited. All rights reserved10

Electron-spinqubit

Superconducting qubit

Magnetic-�eld lines

Microwave resonator

Spin-qubitmemory crystal

m i l e S B l e n C o w e

Most of us have shared the frustra­tion of a desktop computer grind­ing almost to a halt when running a

data­intensive application — opening a high­resolution digital photograph, for example — or running one application too many at the same time. Some have also experienced the (usually short­lived) improvement in speed that comes from installing expensive addi­tional memory called random access memory (RAM). Unlike that in the hard drive, data stored in RAM can be retrieved just as quickly in any order, making it well suited for its role as a temporary storage medium for the computer’s central processing unit while ex ecuting a program. A quantum computer will also require a form of RAM for its proper function. Writing in Physical Review B1 and in Physical Review Letters2–4, four research groups report significant progress in demonstrating a proof­of­principle ‘quantum RAM’.

As with a conventional computer, a quan­tum computer encodes the binary digits 0 and 1 — that is, a ‘bit’ of information — in the state of a physical system. But in contrast to its classical counterpart, it does not do so using the state of a classical system, such as the absence or presence of an electrical charge in a capacitor. The 0 and 1 will correspond instead to the two states of a two­level quantum sys­tem — for example, the spin ‘down’ and spin ‘up’ states of unpaired electrons of atomic or molecular defects in a crystal lattice, or the clockwise and anticlockwise electrical currents in a tiny superconducting ring. And because of quantum mechanics, both of these example quantum systems can be not only in either the 0 or 1 state, but also in a superposition state — a

disease in developed countries — may result from the impaired actions of survival factors such as insulin, in addition to the deleteri­ous effects of increased concentrations of glucose, angiotensin and other factors. This may provide the rationale for evaluating exist­ing insulin­sensitizing drugs, as well as those under development, for their ability to improve insulin action in the kidney. It may also prove fruitful to develop insulin analogues that pref­erentially activate insulin­stimulated mecha­nisms in podocytes and other kidney cells. It is to be hoped that these strategies will provide

further ways to decrease the risk of diabetic nephropathy. ■

Christian Rask-Madsen and George L. King are in the Joslin Diabetes Center, Harvard Medical School, Boston, Massachusetts 02215, USA. e-mail: george.king@joslin.harvard.edu

1. Welsh, G. I. et al. Cell Metab. 12, 329–340 (2010).

2. Breyer, M. D. et al. J. Am. Soc. Nephrol. 16, 27–45 (2005).

3. Kanzaki, M. Endocr. J. 53, 267–293 (2006).

Q U a n T U m C o m P U T i n G

Quantum RAMHybrid quantum systems have been suggested as a potential route to building a quantum computer. The latest research shows that they offer a robust solution to developing a form of random access memory for such a machine.

Figure 1 | Scheme for a hybrid quantum computer. The hybrid quantum­computer architecture envisaged in four papers1–4 combines superconducting qubits and spin qubits. The superconducting­qubit states can be transferred to and retrieved from a spin­qubit memory crystal by microwave­resonator photons. The microwave resonator strongly couples to the crystal’s ensemble of electron­spin memory qubits through the tightly confined oscillating magnetic field of the resonator photons. The resonator photons can also mediate couplings between the superconducting qubits to realize quantum logic gates6.

4. Zelzer, E. et al. EMBO J. 17, 5085–5094 (1998).

5. Eremina, V. et al. J. Clin. Invest. 111, 707–716 (2003).

6. Eremina, V. et al. J. Am. Soc. Nephrol. 17, 724–735 (2006).

7. Chou, E. et al. Circulation 105, 373–379 (2002).8. Fu, Z. & Tindall, D. J. Oncogene 27, 2312–2319

(2008).9. Hermann, C., Assmus, B., Urbich, C., Zeiher, A. M.

& Dimmeler, S. Arterioscler. Thromb. Vasc. Biol. 20, 402–409 (2000).

10. Rask-Madsen, C. & King, G. L. Nature Clin. Pract. Endocrinol. Metab. 3, 46–56 (2007).

11. Tiwari, S., Riazi, S. & Ecelbarger, C. A. Am. J. Physiol. Renal. Physiol. 293, F974–F984 (2007).

simultaneous combination of 0 and 1. As such, they act as a ‘quantum bit’, or qubit, to encode quantum information.

In their papers1–4, the four teams describe work towards realizing a hybrid quantum­computer architecture that combines qubits

of both types described above: superconduct­ing qubits and spin qubits (Fig. 1). The super­conducting qubits, which are typically a few hundred micrometres in size, are well suited to performing fast quantum logic­gate opera­tions5,6. In addition, they are relatively straight­forward to fabricate from materials such as aluminium using electron­beam lithography. The spin qubits, which are formed from a millimetre­to­centimetre­sized crystal con­taining upwards of 1012 electron­spin impu­rities, act as memory elements to store and retrieve data. The spin qubits considered by the authors include chromium(iii) ions (Cr3+) in aluminium oxide2, nitrogen­vacancy defects in diamond3, and rather exotic molecules con­sisting of single nitrogen atoms in fullerene (C60) cages4 (the fullerene cage prevents the normally chemically unstable nitrogen atoms from reacting). Such spin­qubit memories have the advantage of relatively long (mil­lisecond) lifetimes, after which the stored information will typically have decayed away owing to unavoidable interactions with the uncontrolled environment of the spins. This lifetime is more than a thousand times longer than the microsecond or shorter lifetimes of the superconducting qubits.

In such hybrid schemes, the data buses that transport information between compu­ter components can be fashioned from long, thin metallic strips of aluminium or some other suitable superconductor such as nio­bium. The strips can be carefully engineered to form microwave resonators that have resonant frequencies in the few­to­several gigahertz range, enabling the superconduct­ing (logic) and spin (memory) qubits to emit and absorb resonant­frequency microwave photons and hence exchange information between one another.

The new studies1–4 do not demonstrate such a hybrid quantum­computer architecture in its entirety. What the authors show is a strong coupling between the microwave resonator and the ensemble of electron­spin memory qubits in the crystal. To realize a quantum RAM, the coupling must be suf­ficiently strong for a microwave­resonator photon to be stored in the spin­qubit mem­ory and retrieved on a timescale that is short

4 4 | N A T U R E | V O L 4 6 8 | 4 N O V E m b E R 2 0 1 0

NEWS & VIEWSRESEaRch

© 20 Macmillan Publishers Limited. All rights reserved10