Assessing the risk of second malignancies after modern radiotherapy.

The problem of SMNs in paediatric radiotherapy patients is technically complex and challenging. There is good epidemiological evidence that radiation therapy is making a crucial contribution to long-term survival of childhood cancers, but it is also causing a high incidence of SMNs among survivors of childhood cancer. Given the size and consequences of this problem, it is vital that we meet our obligation to ensure that any avoidable and detrimental exposures to radiation are as low as can be reasonably achievable. However, it will be challenging to precisely define a reasonably low level of exposure in terms of risk/benefit, and difficult to predict how much can be reasonably achieved with finite effort and expense 97 . Fortunately, there are several avenues of research available to reduce SMN risks from particle radiotherapy. Modifying treatment units to reduce stray radiation is an obvious and attractive approach 32 . This can be accomplished by reducing the production of stray radiation inside the treatment unit and by increasing its attenuation with additional local shielding. For example, simple enhancements to a passively scattered proton nozzle (shielding) and treatment technique (distance) reduced the dose to paediatric patients by up to 40–50% 95 . New research studies are needed to strengthen the initial evidence used to design personalized treatment strategies, such as the selection of radiation type that is likely to have the least detrimental effects, and treatment techniques, such as beam angles. The long latency times for SMNs require that research uses computational approaches to develop risk mitigation strategies, supplemented with radiobiological and radioepidemiological data as they become available. Moreover, we should compare the stray therapeutic radiation with the dose from diagnostic procedures, which is rapidly increasing and causing concern for long-term risk in the population 98 , 99 . In most cases imaging during therapy accounts for less than 20% of the stray therapeutic radiation, but with the extensive applications of IGRT with daily portal imaging or cone-beam CT using MV X-rays, whole-body exposures of up to 100 mGy per day are possible, thus exceeding those caused by the scatter radiation from the therapeutic beam 100 . When possible, orthogonal pairs of images with kV X-rays should be used to reduce the dose, considering that for paediatric patients the dose to important structures from kV cone-beam CT can be twofold to threefold higher than for adults 101 The main research issues for SMNs in distal organs and for those in the radiation field for paediatric patients are summarized in TABLE 1 . In both high- and low-dose regions, the shape of the dose–response curve and its dependence on fractionation for carcinogenesis should be determined: this is clearly organ-specific, and it is therefore important to collect information about organ doses received during therapy. Even that could not be enough, as in some cases only a part of an organ is exposed to a high dose, and the other part to a low dose. How to deal with such a non-uniform exposure remains an open problem for late effects. Despite the large uncertainties, data collected so far suggest that particle therapy should lead to a lower risk of SMNs than conventional X-ray techniques 102 . This is mainly attributed to lower doses to healthy tissues from the therapeutic beam and to the relatively low risk associated with neutron exposures to the whole body, especially with the spot-scanning delivery method. In other words, less normal tissue is exposed and distal organs receive lower stray radiation. This should also apply to heavy ion therapy, in which the therapeutic dose to healthy tissues is again very low and neutron production even lower than for protons, as fewer particles are needed to achieve the same target dose 30 . In fact, a lower yield of chromosomal aberrations is measured in the peripheral blood lymphocytes of patients treated with C ions compared with X-rays for the same cancers 103 , 104 . Some concerns may arise in the moderate–high dose regions, in which particles can be more effective in inflammation and promotion, although this mechanism may be more relevant for adult patients than for paediatric patients.

In the past few years, enormous progress has been made in radiotherapy 20 – 22 , leading to the possibility of providing escalating radiation doses ( BOX 1 ) to the tumour while sparing the normal surrounding tissue ( FIG. 1 ). This rapid progress has been boosted by advances in imaging tools to identify the tumour, in computerized treatment planning systems and in the development of radiation treatment machines. Contemporary radiotherapy techniques often produce photoneutrons or use accelerated charged particles. Therefore, one of the most important issues is the influence of radiation quality on carcinogenesis. The difference between sparsely and densely ionizing radiation is in the initial energy deposition patterns. The relative biological effectiveness (RBE) of a given radiation is related to the spatial density of energy deposition events, known as the linear energy transfer (LET). A dose delivered by photons (sparsely ionizing or low-LET radiation) is more uniformly distributed than an equal dose of heavy ions (densely ionizing or high-LET radiation). Protons at high energy are similar to X-rays, but at low energy their LET increases and their RBE is higher than α-particles at the same LET 23 . These low-energy secondary protons are responsible for the high RBE of neutrons, arguably the most effective particles in inducing late effects 24 . For heavy ion therapy, target tissues are exposed to carbon ions, and distal organs to neutrons and scattered protons 10 . Because of the unfavourable depth–dose distribution of X-rays, more than one photon beam has to be used to increase the dose to the tumour while sparing the normal surrounding tissue 20 . The improvements in computerized tomography (CT) imaging in the 1980s made it possible to switch to CT-based radiotherapy and three-dimension conformal radiotherapy (3D-CRT). The ability to modulate the intensity of the different beams — intensity-modulated radiation therapy (IMRT) — yielded better target to normal tissue dose ratios (in the moderate–high dose region) and improved sparing of important structures compared with 3D-CRT 25 , 26 . When combined with image-guided radiotherapy (IGRT) technologies, which take into account daily organ motions and tumour shrinkage, IMRT is the ‘state of the art’ photon therapy 27 . However, as first pointed out by Hall and Wuu 28 , the move from 3D-CRT to IMRT has a cost: the risk of SMNs can increase, and may in fact be almost twofold higher. This is essentially caused by two problems: IMRT requires many more fields from different angles than conformal therapy, and therefore a larger volume of normal tissue is exposed to radiation; the exposure of out-of-field tissues from leakage X-rays is also increased, because IMRT requires twofold to threefold more monitor units to deliver a specified dose to the target compared with 3D-CRT. In parallel to advances in photon therapy, radiotherapy with charged particles has evolved substantially since it was originally proposed more than 60 years ago 29 . The rationale of particle therapy is simple: unlike photons, charged particles deposit most of their energy near the end of their range, in the region of the so-called Bragg peak. The Bragg peak is spread out to cover the whole tumour volume, and the exit dose — the dose beyond the tumour — is lower than with photon therapies. Particle therapy can therefore provide excellent dose conformation to the target while simultaneously providing better sparing of normal tissues 10 ( FIG. 1d ). Heavy ions (such as carbon) provide physical advantages compared with photon beams and have a radiobiological benefit. Although high-energy protons have similar LET, and therefore RBE, as photons, carbon ions have low-LET in the entrance region, where the normal tissue is exposed, but high-LET at the end of their range, where the tumour is located 30 . Heavy-ion therapy is therefore preferable for radioresistant cancers, such as sarcomas, renal cell carcinoma, melanoma and glioblastoma 11 . To treat deep-seated tumours with heavy ions, complex treatment systems are required. The size and cost of such systems are presently larger than those of photon therapy; new technologies and economies of scale are expected to lead to smaller, less expensive systems within the next few years. Currently, more than 20 centres are actively treating patients, and over 70,000 cancer patients have received particle therapy 29 . Most of these patients were treated with protons, but the use of heavier ions is increasing 10 . The debate regarding the cost/benefit ratio of this treatment is still ongoing 9 , but many more centres are currently under construction and the number of cancer survivors, especially paediatric patients, treated with charged particles is rapidly increasing 10 . These facts increase the urgency to better understand the risks of SMNs: in fact, neutrons and heavy ions are more effective than photons in cancer induction 31 .

Detailed statistics on the risk of SMNs related to radiotherapy are difficult to obtain because cancer in children is relatively rare. In children, lymphoma, leukaemia, brain tumours, sarcomas, Wilm’s tumour, neuroblastoma and liver cancers are typically treated with radiotherapy 15 . Paediatric patients with tumours in the central nervous system, skull base and bone are those expected to benefit most from treatment with protons 16 or carbon ions 17 . Even though the number of patients treated is fairly small, data obtained so far suggest that proton beam therapy can achieve better local control with no increase in acute toxicity for many childhood solid tumours 18 . Eric J. Hall 12 addressed three major factors that contribute to an increased risk of SMNs in children. First, the atomic bomb survivors database clearly shows that cancer risk depends on the age at exposure and decreases from about 15% per unit dose equivalent (Sv −1 ) at under 10 years of age to about 1% Sv −1 for adults exposed at over 60 years age (see BOX 1 for the definition of dose and Sv). Second, the problem of genetic susceptibility is more serious for children than for adults, in whom carcinogenesis is often associated with lifestyle choices and environmental factors. Many cases of childhood cancer involve a germline mutation, such as the hereditary form of retinoblastoma, and this raises the question of whether the paediatric patient is also more sensitive and susceptible to radiation-induced cancer. Finally, the issue of stray radiation is more serious for children than for adults, simply because of the difference in body size 12 . In its most recent report focused on SMNs, the Childhood Cancer Survivor Study (CCSS) screened 14,359 survivors, and found 2,703 neoplasms 1 . Survivors of Hodgkin’s lymphoma and Ewing’s sarcoma were those at highest SMN risk. A relationship between the relative risk and the dose received for the treatment of the primary cancer could be established for SMNs in the central nervous system, breast and thyroid. Gliomas generally occurred in the high-dose region before an attained age of 30 years in patients who had previously been treated for brain tumours, while breast cancers were observed in low-dose regions after the patients were older than 30 years of age. Taken together, the dose–response relationships, the relative risks and the latency periods for SMNs in the CCSS are consistent with what was expected from the large database of atomic bomb survivors 19 .

Contemporary SMN risk models typically take into account various host-related factors that influence predicted risk, including age at exposure, attained age and sex 62 . Typically, risks are larger for females, decrease with age at exposure and continue to increase with attained age decades after exposure 88 . The subpopulation of patients who are at the greatest risk for the development of SMNs would potentially benefit the most by adding modifying factors to the risk models to take into account lifestyle choices, occupational exposures and genetic variation. To accurately predict radiogenic risk using the models described above, the radiation doses in the tumour, as well as in healthy tissues and organs throughout the entire body, must be calculated. Fortunately, this is now possible, thanks to strident progress in radiation transport models, nuclear interaction data and models, and high-performance computing 14 . This means that, in principle, we can now calculate the dose to the patient and transform it to a cancer-risk map in the body ( FIG. 4 ). In fact, our understanding of the physics of particle therapy is advanced and nearly complete for the purposes of SMN risk assessment. Most of the uncertainty remains in the biology ( TABLE 1 ): we can calculate an organ dose fairly accurately, but the translation of the physical dose into an equivalent dose ( BOX 1 ) and then to an organ-specific risk for cancer incidence and mortality is affected by large uncertainties that are caused by the uncertainties in weighting factors and cancer risk factors. The risk estimates in FIG. 4 are therefore only qualitative, but the calculation with the colour scale clearly shows how the dose is a poor predictor of risk, and large organ-specific differences can be found between incidence and mortality risks. Studies comparing SMN risks that are associated with contemporary proton and photon therapies have consistently found that proton therapy confers smaller overall predicted risk of SMNs for children with medulloblastoma 89 , 90 and for adults with prostate cancer 91 and liver cancer 92 . The models found a negligible difference between lifetime risk of SMN incidence from passive versus scanned proton craniospinal irradiation: the risk was mostly attributable to therapeutic radiation, not leakage neutrons 93 – 95 . Similar findings were obtained for several patients treated for prostate cancer 91 , 96 . Perhaps the most important findings, however, are from comparative studies of passively scattered proton therapy and photon IMRT. These studies revealed that the largest reductions in predicted risk were obtained by replacing IMRT with proton therapy 91 , even after taking into account the stray and leakage neutrons that are associated with proton therapy. In fact, these findings were significant even when taking into account the large uncertainties in neutron RBE values ( FIG. 2c ) for carcinogenesis 90 .

In the past few years, the concerns and controversy about the risk of SMNs following particle therapy has reached unprecedented levels 12 , 32 . This controversy is related to the high cost of the particle therapy facility and the debate regarding the cost/benefit ratio 9 , but it is partly a consequence of a lack of reliable dose and risk assessments tools. Available radiotherapy treatment planning systems do not provide accurate out-of-field dose calculations far from the treatment target 32 – 34 . At present, only a few institutions have the capability to carry out whole-body dose and risk assessments for conventional and advanced radiotherapies 32 , 35 , 36 . Leakage neutrons are a problem for both high-energy photon and charged-particle therapies ( FIG. 2a ). In fact, photons at energies above 8–10 MV (depending on the target material) produce neutrons by photonuclear reactions ( FIG. 2b ). High-energy photons provide a better depth–dose distribution and are therefore preferred over lower energy X-rays for treating deep-seated tumours. High-energy photons interact with materials in collimators and the beam delivery system and can cause photonuclear reactions, which yields unwanted neutrons that contribute to the SMN risk. Photoneutrons are major contributors to doses of 18 MV photons at distances >20 cm from the target 32 , and the dose tends to flatten rather than decrease at greater distance. The patient is therefore immersed in a ‘neutron bath’ ( FIGS 1b , 2a ). With passive proton beam shaping, many neutrons are produced in the treatment unit 32 , 37 , and it has been argued that the leakage of these neutrons may substantially increase the risk of SMNs in distal organs 13 ( FIG. 2 ). Heavy ions also produce secondary particles by nuclear fragmentation, and these particles can deposit their energy in the normal tissue 38 , 39 . Neutrons undergo nuclear collisions with protons in water, generating additional charged particles that can ionize surrounding molecules. Exposures to leakage neutrons can be reduced by using magnetically scanned beams instead of passively scattered beams 30 , 32 , 37 . The production of secondary neutrons in radiotherapy has been measured for different facilities ( FIG. 2b ). A recent review 32 shows substantial inter-centre variability: this variation is attributed to many factors, including differences in the treatment units, experimental techniques and reporting methods. When heavy ions are considered, the neutron yield is caused, in part, by the fragmentation of the 12 C projectiles 39 , which is not relevant in proton therapy. However, this effect is compensated for by the reduced number of 12 C ions necessary to deliver a given dose: approximately 10 10 C ions per treatment, a factor of 100 fewer ions than with proton therapy 30 . Such estimates are affected by very large uncertainties, depending on the specific treatment, site, quality factor and distal organ under consideration. A detailed set of dose measurements for a patient were carried out at GSI Helmholtz Center, Darmstadt, Germany, in 2008, when it was discovered that a young woman treated for a skull-base chordoma with 12 C ions was pregnant 40 . Neutron and photon doses in the uterus region were monitored during the treatment using both active and passive detectors. The dose in the pelvis for the full treatment was 82 μSv, and neutrons contributed about 30% to the total dose. Considering the uncertainty, an upper limit of 0.2 mSv for the uterine dose was estimated 40 . Both the mother and the child are healthy and do not have radiotherapy-related complications 3 years after treatment. Taken together, these data suggest that particle therapy is typically not causing an increase in the dose to distal organs compared with high-energy IMRT. However, great care should be taken in comparing these values, which generally refer to the effective dose 41 , 42 ( BOX 1 ), a controversial radiological unit. Indeed, effective doses use tissue weighting factors that are estimated by several stochastic end points, and do not include any age- or gender-dependence in cancer risk 43 . Especially for paediatric patients, the assumption that weighting factors are independent of age at exposure is tenuous. Cancer is a tissue-specific disease, and there is no evidence that the shape of the dose–response curve is the same for different organs. The International Commission on Radiological Protection (ICRP) indeed recognizes that the use of effective dose in medical applications may be inappropriate and it would be more useful to calculate the risks for specific age and gender groups, using absorbed or equivalent doses to organs and tissues and age-related risk factors 44 .

Ionizing radiation has been recognized as a carcinogenic agent by the World Health Organization for many years 45 , 46 . Although early indications of radiation-induced cancers came from radiologists and other radiation workers, certainly the main epidemiological evidence of radiogenic carcinogenesis in humans and its dose–response relationship comes from the 1945 atomic bomb (A-bomb) survivor cohort 47 . Leukaemia and many solid cancers (especially lung, colon, breast and thyroid cancer) have been linked to radiation exposure 46 . The risks of developing a solid tumour after radiation exposure are reasonably well described by linear dose–response functions in the dose range from 0.2 Sv to 2 Sv ( FIG. 3 ). However, epidemiology does not provide the necessary information for SMNs in radiotherapy patients, in which a small volume is exposed to high doses, and the rest of the body to low doses. The SMN risk is then calculated using models: usually, the linear-no-threshold (LNT) assumption is adopted at low doses 48 , whereas models taking into account competition between cell killing and transformation are used in the high dose region 49 , 50 . Physical doses are converted into dose equivalents using weighting factors at low doses or RBE values at high doses ( BOX 1 ), and then the risk of cancer incidence or mortality is estimated by the product of the equivalent dose and organ-, age- and gender-specific risk coefficients, which have been mostly derived from A-bomb survivors. The models are affected by substantial uncertainties, which can only be reduced with a better understanding of the mechanism of radiogenic carcinogenesis 31 , 48 . What is the molecular mechanism of radiation-induced cancer, especially of therapy-related cancers? Although we do not have a comprehensive answer, it is well known that cancer is a complex multistep process, and radiation can influence both initiation and promotion. DNA damage and repair, particularly double-strand breaks (DSBs), genomic instability and epigenetic mechanisms have key roles 48 . There is no evidence that mechanisms of therapy-related cancers are different from those of sporadic cancers 51 , and therefore SMNs resulting from radiotherapy would just be part of the general problem of radiation-induced carcinogenesis. The relative cancer risks among Japanese A-bomb survivors are generally greater than those among comparable subsets in studies of medically exposed individuals 46 , probably because of the sterilization effects at high doses and the fractionation effects 52 . However, there are a few specific mechanistic issues that are particularly relevant for SMNs in radiotherapy patients: the problem of genetic susceptibility, the possible presence of different mechanisms at low and high doses, and the problems of split doses and radiation quality. Apart from lifestyle factors (such as, smoking and a poor diet), genetic susceptibility is a major confounding factor in determining therapy-related cancers 52 . Can a genetic mutation or polymorphism that is associated with the primary cancer affect the risk and the mechanism of radiation-induced carcinogenesis? Mutations in high-frequency but low-penetrance genes, as well as in low-frequency but high-penetrance genes, can enhance predisposition to radiation-induced cancers. The large increased risk of contralateral second primary breast tumours in women carrying mutations in the high-penetrance genes BRCA1 and BRCA2 provides evidence that mutations in these genes are involved in both the initial cancer and the increased risk of development of the second breast cancer 53 . The other well-known case is ataxia telangiectasia, in which the key gene ataxia telangiectasia mutated ( ATM ) is also an essential signalling molecule for DNA DSB repair following exposure to radiation, and its mutation is linked to genomic instability and cancer 54 . How many genetic defects that are associated with the initial cancer can predispose to a second, radiation-induced malignancy? The recently established Radiogenomics Consortium 5 will carry out genome-wide studies to identify genes that are involved in late effects after exposure to radiation. This is certainly an ambitious programme, especially as a single nucleotide polymorphism (SNP) could be responsible for increased sensitivity. For patients with bilateral retinoblastoma, osteosarcoma is the most frequent SMN that is induced by radiotherapy, which suggests that heterozygosity for the tumour suppressor gene retinoblastoma 1 ( RB1 ) may predispose to radiation-induced bone sarcomas. In fact, genetic mapping in mice showed that multiple loci confer a genetic susceptibility to α-particle-induced osteosarcoma 55 , and Rb1 -heterozygous mice have an increased risk of α-particle-induced osteosarcoma 56 . Interestingly, Ink4a (an upstream regulator of RB1)-heterozygous mice did not have increased susceptibility but had reduced tumour latency after radiation exposure 56 . Animal models and the whole-genome screening in radiotherapy patients are likely to increase our knowledge of the role of genetic predisposition to SMNs. One special characteristic of SMNs following radiotherapy is that they can arise either in the irradiation field (high-dose region) or in the distal organs (low-dose region). The question is whether different biological mechanisms may be causing low dose- and high dose-induced carcinogenesis. The rationale behind this question is that radiation can affect both the initiation and the promotion steps in carcinogenesis. In tissues exposed to low doses, the LNT for the dose–response curve is based on the assumption that even a single electron can induce DNA DSBs that could result in an initial carcinogenic event, even though the probability of this happening is low 57 . At low doses, non-targeted effects 58 may increase the cancer risk by increasing the size of the susceptible target from one single cell to a whole tissue or a part of the tissue, although a protective mechanism has also been hypothesized, in which damaged cells are removed from the organism by intercellular signalling to protect tissue stability. In the high-dose region, ionizing radiation will effectively kill cells in the field. The resulting tissue inflammation can promote carcinogenesis 59 . The response of the microenvironment to radiation and/or oxidative stress is mediated by cytokines, including epidermal and fibroblast growth factors, interleukins and pro-inflammatory cytokines 60 . Activation of the cytokine transforming growth factor-β (TGFβ) is an early and persistent event in tissues that have been exposed to ionizing radiation, and it can have protective or damaging effects: for example, it can protect the stem cell compartment in the intestine, but it can also promote invasion and metastasis in the epithelial tissue 61 . Although epidemiological data clearly show that radiation-induced cancer risk is much higher for exposure during childhood, a recent analysis of the A-bomb survivor cohort suggests that the decrease of cancer risk with age is not continuous for exposure in adulthood 62 . In middle age a number of dormant tumours might be present, so exposure to radiation could cause tumour promotion 63 . Initiation is likely to be the dominant process in the exposure of young people. Therefore, in radiotherapy for paediatric patients, the main concern would be the low-dose exposure of distal organs, whereas for adults high-dose-induced in-field SMNs can be expected owing to inflammation and promotion in pre-neoplastic tissue. Finally, are out-of-field tumours caused by only low-dose, stray radiation? A bystander effect in vivo can be seen after exposure to high doses of radiation, known as an abscopal effect 64 . Recent data have shown an increased incidence of medulloblastoma in Patched 1 ( Ptch1 )-heterozygous mice after high-dose (3 Gy) exposure of the lower body 65 , and apoptosis of unirradiated bone marrow cells can be induced in mice exposed to 4 Gy γ-rays 66 . These high-dose, non-targeted effects are systemic effects, which are transmitted through the nervous, immune or vascular systems, and may be long-range and significant. If radiation at low doses ( FIG. 3a ) does not have an effect on — or even protect against — cancer induction 67 , the non-targeted (abscopal) effect of these high target doses could be responsible for SMNs in distal organs. Radiation quality and its influence on the carcinogenic mechanism is one of the main issues in radiation protection. The radiation weighting factors w R, which are used to calculate the equivalent dose ( BOX 1 ), are among the few parameters that are not based on epidemiology but on laboratory research studies. So far, there is no evidence that radiation carcinogenesis is based on different mechanisms for different radiation qualities 31 . Although high-LET radiation induces complex damage measured both at the DNA 68 , 69 and the chromosomal 70 levels, complex DNA lesions can also be induced by low-LET electrons and are often considered the crucial lesions that lead to late cellular consequences. Non-targeted effects are particularly important in the case of low-fluence high-LET (that is, a low number of particles each delivering a relatively high dose) radiation 71 , but as noted above it is still unclear how much non-targeted effects contribute to radiation carcinogenesis. Gene expression profiles are often very different following exposure to different radiation qualities 72 , and the same is observed switching from low doses to high doses 73 , but gene expression following radiation or any other mutagenic stressor is a complex, time-dependent phenomenon, and difficult to use as a biomarker of late end points such as SMNs. For estimating cancer risk from protons, we must rely entirely on animal and cellular experiments. The US Air Force and NASA carried out one such study from 1963 to 1969 ( REF. 74 ). About 2,000 rhesus monkeys and 5,000 mice were irradiated with protons of energies ranging from 32 MeV to 2,300 MeV obtained using cyclotrons at various institutions. Exposures to electrons and X-rays were also carried out to enable the comparison of the effects of radiation of different qualities. The RBE for protons for acute mortality was about 1.0 to 1.1, which is the RBE generally used in human therapy 75 . The dose–response relationships that could be derived were curvilinear and consistent with those found in other experimental studies of the effects of low-LET radiation. In female monkeys, endometriosis was a major radiation effect that contributed to a shortened lifespan 74 . Another notable finding was the increased incidence of malignant brain tumours in the monkeys exposed to 55 MeV protons. In rodent systems, 250 MeV protons have been shown to be slightly more effective than γ-rays for the induction of Harderian gland 76 or mammary 77 tumours. However, these data are insufficient to provide an estimate of the RBE for cancer induction at the different proton energies. The maximum RBE values for neutrons are observed at low-dose rates, especially for those end points for which a pronounced sparing effect is observed with low-LET radiation. The estimates of neutron relative effectiveness are mostly based on animal studies ( FIG. 3b ), although A-bomb survivors were also exposed to neutrons. In an extensive series of studies at the French Alternative Energies and Atomic Energy Commission (CEA) on male rats, a fission-neutron dose of 20 mGy was found to be equivalent to 1 Gy of acute γ-rays to induce both lethal and non-lethal tumours 78 . Another large database is available from the JANUS programme at the Argonne National Laboratory. In approximately 32,000 mice with acute and fractionated exposures to γ-rays or neutrons 79 , RBE ranged from 2 to 50 and increased with dose fractionation. In these mice, tumours of epithelial tissue were induced by the highest RBE values and tumours of connective tissue origin were induced by the lowest RBE values. Limited data from monkeys on the effects of high-dose total body irradiation showed that the relative risk (that is, the cancer mortality rate relative to mortality rate in unirradiated controls) was approximately 8 for the X-irradiated group and 14 for the neutron-irradiated group 80 . Although RBE for fission-spectrum neutrons has been measured in many experimental systems, less data are available for high-energy neutrons, which are produced in particle therapy 30 ( FIG. 2b ) and in spacecraft 31 . A recent experiment in Medaka fish embryos exposed to a high-energy neutron source at the Los Alamos Neutron Science Center reported an RBE ranging between 25 and 48 relative to γ-rays for apoptosis in cells of the developing brain 81 . Finally, for heavy ions we have probably the highest uncertainty in terms of predicting late effects in humans 31 . Although it is generally assumed that the RBE increases with LET values around 100–200 keV per μm and then decreases at higher values, it depends on several factors, including the tumour type. The RBE of 1 GeV per nucleon 56 Fe ions for inducing cancer in mice was around 1 for leukaemia ( FIG. 3b ) but more than 40 for hepatocellular carcinoma 82 . How can the RBE be so different? The reason is probably due to the different nature of haematological and solid cancers. For haematological cancers radiation could act as an initiator, but for solid cancers (especially those strongly linked to inflammation, such as liver cancer) it could act mainly as a promoter. Although heavy ions are more effective than X-rays in the induction of chromosomal rearrangements, most of the aberrations are lethal, and the RBE for chromosome aberrations drops to about 1 in the surviving cell population 70 . Conversely, heavy ions are effective in the induction of inflammation 83 . In general, only the organs in the beam path are exposed to heavy ions, while distal organs receive neutrons and scattered protons, indicating that SMNs in heavy-ion therapy could be most relevant to adult patients. A further complication in radiotherapy is caused by an incomplete knowledge of the biological effects of dose fractionation. In classical radiobiology textbooks it is stated that the same dose delivered at a low dose rate or in daily fractions is less effective than acute exposure (sparing effect), but the extrapolation from high- to low-dose rate is one of the main sources of uncertainty in radiogenic cancer risk estimates 84 . In a recent, careful review of available experimental data (including human, animal and in vitro data) on radiation carcinogenesis, Suit et al. 85 concluded that fractionation does not necessarily lead to a reduced cancer risk. As some SMNs could be caused by exposure to low-dose neutrons, it should also be noted that some in vitro data suggested that an inverse dose-rate effect exists for neutrons — carcinogenesis would be enhanced at a low-dose rate compared with acute exposure to fission-spectrum neutrons 86 . The inverse dose-rate effect is controversial, it has not been confirmed in subsequent experiments, and it might be visible and relevant only in a small window of dose rates, doses and LET 87 . The issue of carcinogenicity of split doses also remains unclear both in high- and low-dose regions, and more experiments to clarify this issue are urgently needed.