Analysis of amyotrophic lateral sclerosis as a multistep process: a population-based modelling study.

BACKGROUND: Amyotrophic lateral sclerosis shares characteristics with some cancers, such as onset being more common in later life, progression usually being rapid, the disease affecting a particular cell type, and showing complex inheritance. We used a model originally applied to cancer epidemiology to investigate the hypothesis that amyotrophic lateral sclerosis is a multistep process.
METHODS: We generated incidence data by age and sex from amyotrophic lateral sclerosis population registers in Ireland (registration dates 1995-2012), the Netherlands (2006-12), Italy (1995-2004), Scotland (1989-98), and England (2002-09), and calculated age and sex-adjusted incidences for each register. We regressed the log of age-specific incidence against the log of age with least squares regression. We did the analyses within each register, and also did a combined analysis, adjusting for register.
FINDINGS: We identified 6274 cases of amyotrophic lateral sclerosis from a catchment population of about 34 million people. We noted a linear relationship between log incidence and log age in all five registers: England r(2)=0·95, Ireland r(2)=0·99, Italy r(2)=0·95, the Netherlands r(2)=0·99, and Scotland r(2)=0·97; overall r(2)=0·99. All five registers gave similar estimates of the linear slope ranging from 4·5 to 5·1, with overlapping confidence intervals. The combination of all five registers gave an overall slope of 4·8 (95% CI 4·5-5·0), with similar estimates for men (4·6, 4·3-4·9) and women (5·0, 4·5-5·5).
INTERPRETATION: A linear relationship between the log incidence and log age of onset of amyotrophic lateral sclerosis is consistent with a multistage model of disease. The slope estimate suggests that amyotrophic lateral sclerosis is a six-step process. Identification of these steps could lead to preventive and therapeutic avenues. FUNDING: UK Medical Research Council; UK Economic and Social Research Council; Ireland Health Research Board; The Netherlands Organisation for Health Research and Development (ZonMw); the Ministry of Health and Ministry of Education, University, and Research in Italy; the Motor Neurone Disease Association of England, Wales, and Northern Ireland; and the European Commission (Seventh Framework Programme).

Introduction

Amyotrophic lateral sclerosis is a neurodegenerative disease that mainly affects upper and lower motor neurons. It shows complex inheritance: about 5% of people with amyotrophic lateral sclerosis have a family history of the disease or frontotemporal dementia in a first degree relative and up to 20% have an affected relative in more extensive population-based family studies.

Amyotrophic lateral sclerosis has several intriguing features (many shared with other neurodegenerative diseases) that remain unexplained. First, amyotrophic lateral sclerosis is an adult-onset disorder, even in individuals born with a gene mutation that increases the risk of the disease. Although such a mutation is carried from birth, many people remain healthy into old age and do not develop the disease. Others remain completely healthy until disease onset seems to begin suddenly (typically between the age of 50 and 70 years), and progresses rapidly. It is unknown why a pathological genetic change present from birth is expressed only in adult life, in some people but not others, even for high-penetrance mutations (eg, in the SOD1 gene), and why, when the mutation is expressed, the pathological process progresses rapidly. Furthermore, several amyotrophic lateral sclerosis genes show pleiotropy, in which the same gene mutation can result in different phenotypes. For example, individuals with expansion of a hexanucleotide repeat in the C9orf72 gene can remain healthy, or might develop amyotrophic lateral sclerosis, frontotemporal dementia, or amyotrophic lateral sclerosis–frontotemporal dementia. The same mutation might also predispose to schizophrenia, depression, and Parkinson's disease but in every case the burden seems to be specific to a particular subgroup of cells. Additionally, amyotrophic lateral sclerosis seems to start in one neural region and spread, but no genetic or environmental factor has yet been found to decide the site of onset.

Several of these characteristics are shared with cancer, which suggests that, despite the differences between cancer and neurodegeneration (eg, cancer is an uncontrolled proliferation of cells, whereas neurodegeneration is the result of the death of cells), other shared features remain to be discovered.

Since the 1950s, multistep models have been applied to the study of population patterns of cancer and, although the level of mathematical support remains a matter of debate, they have yielded insights into the likely causes of cancer and in some cases the identification of the steps involved. These models generally show that a plot of epithelial cancer incidence against age has an exponential pattern; incidence is proportional to age raised to the power six. This association is consistent with the hypothesis that these cancers are the end result of seven successive mutations. Different patterns are reported with some specific cancers—eg, breast cancer, in which cell replication at particular stages of life might have an important role.

Like cancer, amyotrophic lateral sclerosis might be a multistep process in which several sequential steps are needed. For example, a high-penetrance disease-causing mutation would still need the accumulation of the remaining steps to result in disease, which would take time. This scenario would explain both the adult onset and the finding that not every individual carrying a mutation develops disease.

Multistep models have not been used previously in the study of neurodegenerative disease. Therefore, we used a model originally applied to cancer epidemiology to test the hypothesis that amyotrophic lateral sclerosis is a multistep process.

Methods

Model

We used the approach outlined by Armitage and Doll. Briefly, if one assumes that amyotrophic lateral sclerosis is caused in one step, the incidence in a particular year, i, will be proportional to the risk of having undergone that specific step in that year; this risk, u, will depend on the level of exposure to the relevant disease-causing factor. However, without knowledge of that factor and the exposure level, the incidence will be proportional to the average background risk, u, of this step. If instead, the disease needs more than one step (each step with risk ui), then the chance of undergoing the first step by age t years is u1t. Undergoing the second step by age t has risk u2t, and so on, until the state is reached after n–1 steps, in which the person is primed so that the next and final step, which has risk u, would result in disease. Each risk u is assumed to be small because an exposure effect with probability close to 1 makes no difference to the product.

Therefore:

and

log(

(where c is a constant representing log(u1u2u3...uu)

The model can be modified to allow for the assumption that the changes need to happen in a specific order, but this assumption does not alter the age-incidence patterns predicted by the model.

Because i is incidence and t is age, a plot of the log of amyotrophic lateral sclerosis incidence against the log of age will be linear if a multistep model applies, and will have slope n–1, one less than the number of steps needed. One exception should be noted: the model predicts that the slope will decrease (and therefore will be less than linear) at the older age groups, a pattern that has been recorded in cancer.

Population registers

To identify population incidence data for amyotrophic lateral sclerosis, we searched PubMed and contacted amyotrophic lateral sclerosis epidemiology research groups. We generated incidence data by age and sex from five amyotrophic lateral sclerosis population registers, which provided incidence and prevalence data broken down by sex and 5-year age groups in Ireland, the Netherlands, Italy (Piedmont), Scotland, and England (South East England amyotrophic lateral sclerosis register [SEALS]). All participants in the registers had provided written consent for inclusion and subsequent analysis of their data. All five registers try to capture every incident case of amyotrophic lateral sclerosis within a defined catchment area over several years, which allows age and sex-adjusted incidence to be estimated.

To provide a comparison with another neurological disease, we examined population-level data for multiple sclerosis. Data were obtained from a study from Manitoba, Canada.

Statistical analysis

For each register, we used the 2013 version of the European standard population to calculate age-standardised incidence rates for the 25–74 years age range per 100 000 person-years with 5-year age groups (table 1).

Table 1

Basic demographic features of each population register

	Registration dates	Age-standardised incidences (men, women, both)	Number registered	Catchment population size	Missing data (n)	Mean age of onset, years (SD)	Number of men (%)	Number with bulbar onset (%)
England (SEALS)	2002–09	2·19, 1·48, 1·84	438	4·023 million	27	62·1 (12·0)	228 (55%)	131 (32%)
Ireland	1995–2012	3·27, 2·38, 2·82	1593	4·093 million	239	65·1 (11·8)	771 (57%)	486 (36%)
Italy (Piedmont)	1995–2004	3·19, 2·20, 2·69	1260	4·398 million	0	64·8 (11·2)	687 (55%)	473 (38%)
Netherlands	2006–12	2·16, 1·48, 1·82	1757	16·546 million	66	63·4 (11·4)	1017 (58%)	589 (35%)
Scotland	1989–98	3·43, 2·23, 2·83	1226	5·125 million	61	66·3 (11·0)	666 (54%)	430 (37%)

For the Netherlands and Scotland, no patients had missing age of onset data; for these countries, missing data refers to site of onset of first symptoms (bulbar or spinal). Therefore, the age of onset, numbers and proportions of men, and incidence calculations used the full set of patients. For England and Ireland, missing age of onset is reported; all patients missing site of onset were also missing age of onset (two patients in England and eight in Ireland). There were no missing data for the Italian register. SEALS=South East England Amyotrophic Lateral Sclerosis register.

We did two different analyses: first, we included data from all ten 5-year age groups from 25–29 years through to 70–74 years; and second, we omitted the youngest and the oldest age groups. We did this because analyses of amyotrophic lateral sclerosis incidence by age, as in cancer, might be imprecise and have measurement errors in extreme age groups. In particular, rates in young people might be imprecise because of the small numbers of cases. Rates in old age groups might be underestimated because of case under-ascertainment or cohort effects (ie, generation differences); for example, this pattern has been recorded for lung cancer incidence because older people had grown up during a period when smoking rates were relatively low. Alternatively, some versions of the multistep model predict that the log(incidence) and log(age) association might be less than linear in the older age groups.

We estimated age and sex-adjusted incidence rates separately for each population register, in 5-year age groups. We used the midpoint of each age group in the regression analyses. We regressed the log of the incidence rate against the log of age with linear least squares regression. We used unweighted regressions so that the slope was not biased towards that of the oldest age groups (which included the most people), but we also checked our results with a Poisson regression, which gave greater weight to age groups that had the largest numbers of cases. We did the analyses within each centre, and also did a combined analysis adjusting for centre. All analyses were done in STATA (version 12). The regression used was the unweighted linear least squares regression using all age groups.

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. All authors had full access to all the data in the study. The corresponding author had final responsibility for the decision to submit for publication.

Results

The five amyotrophic lateral sclerosis population registers gave similar values for the overall incidence of amyotrophic lateral sclerosis and within each age group. Table 1 shows results for each population register.

There was a linear relationship between log incidence and log age in all five population registers (figure 1, appendix): England r2=0·95, Ireland r2=0·99, Italy r2=0·95, the Netherlands r2=0·99, Scotland r2=0·97, and overall r2=0·99, entirely consistent with a multistep model.

Figure 1

Log(incidence) vs log(age) for the five registers combined

The fit to a straight line is good (r=0·99), consistent with a multistep model.

All five registers gave very similar estimates of n–1 ranging from 4·5 to 5·1, with overlapping confidence intervals (table 2 and figure 2). When combined, the five registers gave an overall slope of 4·8 (95% CI 4·5–5·0) with similar values for men (4·6, 4·3–4·9) and women (5·0, 4·5–5·5). Excluding extreme age groups with low counts made little difference to the results (overall slope 5·1, 4·8–5·4). A weighted Poisson regression gave slightly lower slopes, showing the sublinear patterns in the older age groups that have more weight in these analyses (overall slope 4·7, 4·5–4·8).

Table 2

Estimates of slope from each amyotrophic lateral sclerosis register (all age groups [25–74 years])

	England (SEALS)	Ireland	Italy (Piedmont)	Netherlands	Scotland	Total
Total	4·7 (3·9–5·5)	5·1 (4·7–5·4)	4·5 (3·8–5·2)	4·6 (4·2–5·0)	5·0 (4·3–5·6)	4·8 (4·5–5·0)
Men	3·5 (2·7–4·4)	5·0 (4·4–5·6)	5·0 (4·6–5·5)	4·3 (3·9–4·6)	4·8 (4·3–5·3)	4·6 (4·3–4·9)
Women	4·7 (3·9–5·5)	5·5 (4·7–6·3)	4·3 (2·6–6·1)	5·2 (4·6–5·8)	5·3 (4·2–6·5)	5·0 (4·5–5·5)

Data are slope (n−1; 95% CI). SEALS=South East England Amyotrophic Lateral Sclerosis register.

Figure 2

Forest plot of the estimates of n–1 from each register and overall in all age groups with a weighted average across registers

All registers had results consistent with a slope (n–1) estimate of 5, and the meta-analysis showed little evidence of heterogeneity. The weights of each study shown by the size of the grey squares are England 7·0%, Ireland 38·0%, Italy 9·8%, the Netherlands 34·4%, and Scotland 10·8%. The black diamonds show the estimate of n–1, with the bars representing the 95% CI.

Figure 3 shows findings for multiple sclerosis. The absence of a straight line suggests that an alternative model, such as simultaneous multiple hits, dose-dependent exposure, or continuum of risk operates.

Figure 3

Log(incidence) vs log(age) for multiple sclerosis

Data from the Manitoba register. The line is not straight, showing that the disease process does not fit to a multistep model.

Discussion

We noted a linear relationship between the log incidence and log age of onset of amyotrophic lateral sclerosis, consistent with a multistep model of disease. This association is also consistent with other models, but our findings cannot be taken as definitive proof that amyotrophic lateral sclerosis results from a multistep process (panel). However, our findings provide support for this initial hypothesis, and suggest that it should be explored further. If the model is correct, then the slope estimate is about 5, which suggests that amyotrophic lateral sclerosis is a six-step process.

The findings are similar across all five populations studied, and in both sexes, suggesting that similar causal mechanisms for amyotrophic lateral sclerosis exist across western European populations. The slope is not exactly five in all five registers, but this is to be expected, because variations in slope might arise by chance and because of cohort effects, in which people born in different time periods have different risks at the same age because of differences in exposure to environmental factors. Our results do not prove that there are six steps in each individual, and some people might have fewer or more steps, but the findings in the five registers are consistent with an overall slope of about five. The similarity of the findings in five independent population registers increases our confidence in the reliability of these results. The use of population registers is a further strength of the study design because it minimises bias through almost complete case ascertainment. Furthermore, our finding is not simply an artifact that would arise irrespective of the disease under investigation because we did not find a linear relationship in multiple sclerosis. However, other neurodegenerative diseases might also follow a linear relationship consistent with a multistep model.

If amyotrophic lateral sclerosis results from a multistep process, there are profound implications for our understanding of the disease mechanisms and the interplay of risk factors. The finding of large-effect mendelian gene variations that seem sufficient to cause amyotrophic lateral sclerosis, such as those in SOD1, TARDBP, FUS, or C9orf72, might suggest a one-step process in some individuals. However, such an interpretation does not explain why disease onset arises in adult life, or in some cases not at all. Various models have been proposed concluding that the pathological process is present from birth, but that the toxic effect of the disease-causing protein takes time to build up to sufficient levels to trigger disease. Although this explanation might seem attractive, there is no evidence for it, and it fails to explain genetic pleiotropy, the great variability of onset age even within the same family, and why the disease process then cascades across the motor system so quickly. In most people, amyotrophic lateral sclerosis arises as an apparently sporadic process, which suggests the role of many small-effect genetic factors and their interactions with environmental risk factors. Therefore, a commonly used model for amyotrophic lateral sclerosis is a dose-dependent, two-step process of genetic risk and subsequent environmental triggers.

Our model is at odds with both these interpretations and suggests that six genetic or environmental exposures are needed, and that the last one (which could be environmental or genetic because gene expression varies through life) is the disease trigger. This finding means that environmental studies should be done in all individuals with amyotrophic lateral sclerosis, including those with familial disease, and that identification of the environmental risk factors could lead to a preventive strategy for those carrying a disease mutation who are presently unaffected. One method to explore the nature of these subsequent steps would be an analysis of pooled cohorts that differ in fundamental exposures, such as smoking habits or alcohol consumption. Alternatively, individuals with large-effect mendelian gene mutations would be expected to have inherited one of the steps and should therefore show a slope consistent with five further steps.

The multistep model is consistent with the abiotrophy hypothesis of amyotrophic lateral sclerosis pathogenesis, which posits that a toxic insult specifically depletes motor neurons: the toxic insult would be one of the steps, either early on, with subsequent steps needed to damage remaining neurons, or late, with only neurons that had been through the earlier steps being susceptible to the insult. The multistep model is also consistent with a recently proposed model of neurodegenerative diseases being the result of somatic mutations. In the somatic mutation model, the huge number of cell divisions needed to generate an adult human being inevitably results in the accumulation of somatic mutations, with the possibility of the acquisition of mutations that predispose to neurodegeneration in the nervous system. This model is very similar to present models of cancer in which a subset of cells becomes precancerous through somatic mutation, as one of the steps needed for carcinogenesis. The timing of any somatic mutation, for example, at the fourth rather than the ninth cell division, makes a huge difference in the proportion of the billions of cells carrying a mutation. In a multistep process, the cell-averaged rate of the next hit leads to large differences in projected ultimate risk. A consequence is that the probability of disease for some individuals is close to zero, and for others very high.

A multistep model is also consistent with present hypotheses that amyotrophic lateral sclerosis is the result of long-term aggregation of protein or seeding of prion-like domains in cellular proteins, because one of the steps could include protein aggregation in a subset of cells, and a further step could include failure to clear the toxic aggregates in a smaller subset. Thus the site of onset can be explained because only some neurons will have undergone all necessary steps.

A multistep model means that some crucial environmental exposures might have arisen in the distant past (ie, early in life), which makes them difficult to identify. Furthermore, different agents might cause the same step to happen, which suggests there are many alternative possible pathway combinations. This scenario would mean that some environmental factors are relevant in only a subset of people, which also reduces the statistical power to identify such associations.

There is increasing evidence that cancer and neurodegeneration are epidemiologically and genetically linked. The finding that both can be modelled as a multistep process further strengthens this association.

Although data from some studies suggest mutant SOD1-mediated familial amyotrophic lateral sclerosis has a mean age of onset about 10 years younger than apparently sporadic amyotrophic lateral sclerosis, this finding has not been confirmed in a population-based study, or for familial amyotrophic lateral sclerosis. One explanation is ascertainment bias. However, findings of a population study have shown that people with amyotrophic lateral sclerosis carrying a disease variant of C9orf72 have a significantly younger age of onset than those without such variants. Thus, individuals with mutations might need fewer subsequent risk exposures. Analysis of this group might independently show a shallow slope; in particular, individuals with a mutation that accounts for one step might be expected to have a log(age) versus log(incidence) slope of 4 rather than 5. Further studies are needed to assess this issue. Future work should also analyse other subgroups, such as those with bulbar onset disease. Our model allows that some steps might be the same across different neurological disorders, and the exact sequence or nature of the exposures needed to induce the relevant steps would mean that different clinical patterns arise, thus explaining genetic pleiotropy.

This study has limitations. The population registers used were all based in western Europe; therefore, the findings might not generalise to populations outside this region. Although our data are consistent with a multistep model, other models could result in a similar association between age and incidence. For example, if amyotrophic lateral sclerosis was the result of one step happening in each of a cluster of six cells, followed by prion-like spread, then the same relationship might be recorded, although only one step has happened in each cell.

Our findings support the hypothesis that a multistep process requiring just six distinct steps leads to the onset of amyotrophic lateral sclerosis. This provides hope that the identification of the steps could therefore lead to preventive or therapeutic avenues that have not yet been examined. The consistency of our findings across five western European populations and in both sexes suggests that the underlying mechanism is similar, and increases our confidence in the validity of the results. This hypothesis provides an explanation for several intriguing features of amyotrophic lateral sclerosis and means that environmental studies are essential even in people at high risk because of inherited genetic susceptibility.