July 07, 2010
New Autism Consortium Study Proves (Again) that Inherited Genes Don’t Cause Autism
By Mark BlaxillThe Autism Genome Project Consortium (AGPC) deserves credit for at least one thing. They know how to make a public relations mountain out of a scientific molehill.
The AGPC was all over the news last month, leading a celebration that went global. “Genes breakthrough in autism study” proclaimed a UK headline. “Genetic link to autism found”, announced the Sydney Morning Herald, while the Los Angeles Times told us with only bit more circumspection that a “Big study identifies new genes that may be involved in autism.” The publicity blitz included the suggestion of practical applications, with a report from the Daily Mirror in London promising that an “Autism find could lead to a new test” whileBusiness Week picked up more directly on the commercial opportunity, “Autism’s genetic ties may lead to early detection, treatment.”
We’ve seen this movie before. No really, not just the metaphorical repetition of the autism genetics script, but a real movie. It’s calledGroundhog Day. Like Bill Murray’s character waking up every morning on the same day, the autism gene hunters relive this same public relations routine over and over again with full awareness of their past failures. Meanwhile the media, like Murray’s fellow travelers in the movie, remain blessedly free of any long term memory and go willingly along for every ride.
Genetic analysis once again finds…not very much
So what, you might ask, was all the excitement about? In brief, the AGPC reported evidence that a particular type of rare, inherited mutation (called a CNV, or copy number variant) was more frequent in the DNA of autistic cases than controls. Unlike past autism-CNV studies, the authors did not report that the total number of mutations per individual was more frequent in autism. Instead, according to the AGPC, the specific mutations found in autism cases were more frequently located on active genes. In the eyes of the study authors, this “higher global burden of rare, genic CNVs” suggests that the location rather than the frequency of CNVs on the autistic genome made them likely to be disruptive in their biological effect.
One thing to know about CNVs is that they come in two flavors. They can include either missing DNA—deletions, or extra stretches of a DNA sequence—duplications. The AGPC study attempted to tie both flavors to autism. In the AGPC account, however, the coherence of the argument around these two CNV flavors mattered far less than their ability to report encouraging progress in the decades-long hunt for autism genes.
Not surprisingly, there was little encouraging news for families affected by autism in the fine print, nor was there much clarity over the CNV specifics. Notably, the AGPC authors couldn’t quite pinpoint which flavor of CNV--deletions or duplications-- was at the root of the problem in autism. Across all genes, they calculated that the DNA of autistic cases harbored more deletions but not duplications; in terms of the subset of so-called “ASD-implicated” genes (more on these below), the deletions weren’t the problem it was the duplications; yet when it came down to extracting causal insight from all these CNVs, duplications were out and deletions were in.
So here’s a simple question for the AGPC. Which is it, are deletions or duplications associated with autism?
Well, and again not surprisingly, the answer depends on what argument they’re trying to make. And the flip from deletions to duplications and back again reveals the AGPC’s progressive retreat, a retreat away from objective findings and towards questionable and therefore manipulable constructs.
The most compelling part of their evidence, the finding that is featured most actively, points to the deletion evidence. This finding emerges from the AGPC’s exhaustive analysis examining how frequently CNVs show up on the autism genome relative to unaffected controls. They report their single positive finding with great fanfare, but if one reads the study carefully (including the 74 pages of supplemental materials), one learns quickly that the vast majority of the results are actually negative. Most astonishing is that there was essentially no difference at all in the mutation rate between autistic subjects and controls, a finding that goes against several prior studies. In fact, along just about every dimensions, the AGPC team found no difference in CNVs between cases and controls: no excess of CNVs in terms of the number of CNVs per sample, the percentage of samples with CNVs, the total amount of affected DNA, or the average size of CNVs. This was true for deletions and duplications and for small and large CNVs. Nor were there any differences between families with a single child (simplex) or multiple affected (multiplex) children.
The single finding of significance relied on specifying the location of the CNVs in autism cases, i.e. that they were on active genes, the small fraction of the genome that actually encodes for proteins. And here, the AGPC found that the autistic genome more frequently demonstrated stretches of lost DNA—deletions--on such genes. This is an interesting, if hardly revolutionary, finding but one that goes essentially nowhere in explaining causation. Even though the suggestion of a different mutation rate of just the genic regions of the autism genome narrows the range of biology under inspection, we’re still talking about the entire set of protein-coding human genes, roughly 21,000 of them according to recent estimates.
So, although the suggestion of an elevated rate of missing genic DNA in autism is interesting, the AGPC obviously needed to look for something more specific. And here, their deletion evidence failed them, because although there appeared to be more deletions in allgenic regions, there weren’t really any differences in deletion rates on the genes of highest interest. Over 170 orthodox autism researchers were listed as co-authors on the study. For years, these researchers have been working overtime to specify “autism genes.” And over that period they’ve accumulated long lists of faint signals (rarely replicated with any consistency and often abandoned): genes they believe they’ve connected to autism. But when it came time to zoom in on their putative “global burden of rare, genic” deletions, the AGPC wasn’t able to connect that burden of deletions to the long list of genes they’ve been trying to implicate in autism for so long.
In fact, upon closer inspection, the deletion findings were downright embarrassing for the authors. Deletion rates in cases and controls were virtually identical on all kinds of genes that have been connected to autism. So despite a purportedly “breakthrough” finding on of an increased rate of missing genic DNA in autism, the authors had to turn to the other flavor of CNVs—duplications--to salvage some kind of biological case from their data.
For this purpose, they turned to a second analysis based on three different short lists of genes:
1) Markers of genetic syndromes like Fragile X, neurofibromatosis, Smith-Lemli-Opitz and tuberous sclerosis that are known to increase susceptibility to autism (“ASD-implicated”);
2) Genes that have been suggested in previous autism research (“ASD candidates”); and
3) Genes that have never been connected to autism but are involved in other intellectual disabilities (“ID”).
In this second analysis, they reported a small, but significant increase in the duplication rate in “ASD implicated” genes. That’s about it. (Neither deletions nor duplications showed up significant on the ID gene list, but when added together a small ID effect was just barely significant).
1) Markers of genetic syndromes like Fragile X, neurofibromatosis, Smith-Lemli-Opitz and tuberous sclerosis that are known to increase susceptibility to autism (“ASD-implicated”);
2) Genes that have been suggested in previous autism research (“ASD candidates”); and
3) Genes that have never been connected to autism but are involved in other intellectual disabilities (“ID”).
In this second analysis, they reported a small, but significant increase in the duplication rate in “ASD implicated” genes. That’s about it. (Neither deletions nor duplications showed up significant on the ID gene list, but when added together a small ID effect was just barely significant).
But in a shocking setback for decades of autism research, the findings on over 100 “ASD candidate” genes were completely negative. Anyone with fond Groundhog Day memories of earlier publicity blitzes for the HOXA1 gene, Cadherin 10, the serotonin transporter gene? There’s not even a hint of a positive in this long list of previously ballyhooed genes that made up the AGPC’s “ASD candidate list.”
What’s more, although the AGPC’s duplication findings were positive, their effect was trivial. For the only significant finding--duplications on the ASD-implicated genic regions--the “population attributable risk” (PAR) came out to less than 2%. Under the best possible scenario (one in which they take credit for individually insignificant effects in ID deletions and duplications and ASD-Implicated deletions), the PAR was 3.3%.
Of course, the headlines never read, “after years of effort, a multi-million dollar AGPC study connects gene mutations to less than 2% of autism cases.” But the raw truth in the fine print reveals that this weak finding has no right to be called a “breakthrough”: most of the genetic defects involved in these ASD or ID implicated syndromes are vanishingly rare; none of them “cause” autism, they merely create susceptibility; and the suggested biological mechanisms are as diverse as the many syndromes lumped together as “ASD-implicated” and “ID.” This is just the kind of constructed, tortured finding that falls apart on closer inspection. Will this molehill of a scientific finding ever be replicated? We’ll have to wait and see.
Moving beyond their second analysis based on these somewhat arbitrary lists, the researchers turned to a third approach in order to investigate possible biological mechanisms in their faint CNV signal. This last approach looked at cluster of genes that have been assembled by other groups into “gene sets” (clusters of between 5 to 700 genes) based on known metabolic connections between the proteins these genes sets produce. In this last analysis the AGPC researchers asked whether the mutations they saw in their autism cases showed any tendency to cluster in certain subsets of connected genes.
In this third go around, the AGPC team turned away from their brief infatuation with duplications and returned to deletions. In their analysis of how the mutations clustered in identifiable biological pathways, “only deletions were found to be significantly enriched in gene sets in cases over controls.”
So was there anything of value that might come from this gene set finding, any insights on the altered metabolism of autism that might provide clues to its cause, prevention or treatment? Reporting their findings without much interpretation, the authors merely offered the hope that their work might “lead to final connected pathways.” Let’s hope that some good may ultimately be extracted from all this investment. But is it even remotely conceivable that any of these pathways supposedly “enriched in deletions” going to lead us to viable insight about prevention or targets for treatment of autism anytime soon?
I wouldn’t hold your breath.
The devil in the details
So far in this critique, I’ve made the assumption that, hype aside, the basic evidence in the AGPC paper is reliable as far as it goes. After all, with dozens of authors, presumably rigorous peer review and the imprimatur of the prestigious journal Nature, one can at least expect to rely on the integrity of the methods the AGPC authors used and the validity of the results obtained thereby. Right?
Well…
Although, I wouldn’t make the case that Nature has fallen to the level of Pediatrics in terms of de minimus quality standards (ahhPediatrics! the refuge of such airtight study designs as Verstraeten et al, 2003, Madsen/Thorsen et al, 2003 and Fombonne, 2000) it’s hard not to raise your eyebrows at a number of aspects of the AGPC’s study design. At the most basic level, a few notable concerns jump out.
• The controls aren’t matched to the cases. It’s a fundamental principle of any quality case-control study that the case group should resemble the control group as closely as possible. Here, and oddly, the demographic profile of the AGPC’s autism subjects is not reported at all.
Yet it’s almost certain that their case population differs sharply from controls, a profile of which is provided. For one thing, the control population is 69% female, while the case population is almost certainly disproportionately male. In addition, this heavily female group was, on average, 39 years old at the time of sample collection; the age of the autism subjects was almost certainly younger than that. This profound mismatch of gender and age in the study population raises all kinds of interesting questions. What if the female controls were less prone to mutagenic damage and carried lower mutation rates? What if birth date is a key risk factor for CNVs? What if some other unreported population mismatch underlies the faint difference in CNVs the group reported? Suffice it to say, when cases and controls aren’t rigorously matched, any report of a significant difference between the two groups is suspect.
• The autism samples are thrown out more often. As in all CNV research, quality control issues abound when it comes to detecting the minute changes in DNA that qualify as deletions or duplications. Indeed, one need only scan the methods sections of any CNV study to discern the importance of quality control in the interpretation of CNV “calls.” In the case of the AGPC report, despite manifestly rigorous efforts to ensure the reliability of their CNV findings, the possibility remains that the results are simply artifacts. Why? Because the case and control samples come from completely different sources: the autism samples from the various consortium members and the control samples from the Study on Addiction, Genetics, and the Environment. And were there differences in the handling of samples between these two sources? It’s hard to know but certainly possible. If one reads the methods section closely, one can calculate that the rate of autism samples thrown out for quality reasons was nearly twice (11.2%) as high as the rejection rate for controls (6.2%).
• Duplicate reporting perpetuates possibly random findings. There’s one issue that’s hard to sort out from the methods described: the possibility of duplicate reporting of results within the autism group. The AGPC autism sample relies at least in part autism cases derived from the AGRE repository. The AGRE cases have been deployed multiple times, and at least some of the replication of CNV findings in the AGPC report occurs because the same samples are included in more than one study.
• The controls aren’t matched to the cases. It’s a fundamental principle of any quality case-control study that the case group should resemble the control group as closely as possible. Here, and oddly, the demographic profile of the AGPC’s autism subjects is not reported at all.
Yet it’s almost certain that their case population differs sharply from controls, a profile of which is provided. For one thing, the control population is 69% female, while the case population is almost certainly disproportionately male. In addition, this heavily female group was, on average, 39 years old at the time of sample collection; the age of the autism subjects was almost certainly younger than that. This profound mismatch of gender and age in the study population raises all kinds of interesting questions. What if the female controls were less prone to mutagenic damage and carried lower mutation rates? What if birth date is a key risk factor for CNVs? What if some other unreported population mismatch underlies the faint difference in CNVs the group reported? Suffice it to say, when cases and controls aren’t rigorously matched, any report of a significant difference between the two groups is suspect.
• The autism samples are thrown out more often. As in all CNV research, quality control issues abound when it comes to detecting the minute changes in DNA that qualify as deletions or duplications. Indeed, one need only scan the methods sections of any CNV study to discern the importance of quality control in the interpretation of CNV “calls.” In the case of the AGPC report, despite manifestly rigorous efforts to ensure the reliability of their CNV findings, the possibility remains that the results are simply artifacts. Why? Because the case and control samples come from completely different sources: the autism samples from the various consortium members and the control samples from the Study on Addiction, Genetics, and the Environment. And were there differences in the handling of samples between these two sources? It’s hard to know but certainly possible. If one reads the methods section closely, one can calculate that the rate of autism samples thrown out for quality reasons was nearly twice (11.2%) as high as the rejection rate for controls (6.2%).
• Duplicate reporting perpetuates possibly random findings. There’s one issue that’s hard to sort out from the methods described: the possibility of duplicate reporting of results within the autism group. The AGPC autism sample relies at least in part autism cases derived from the AGRE repository. The AGRE cases have been deployed multiple times, and at least some of the replication of CNV findings in the AGPC report occurs because the same samples are included in more than one study.
More important than these three basic concerns, however, is the possibility of more subtle design errors, especially in the parts of the study where the authors have developed original conceptual frameworks in their quest for biological relevance. For example, a key part of the AGPC analysis relies on a bucketing of genes into three categories: “ASD-implicated” genes, “ASD candidate” genes and “intellectual disability” genes. But these are categories created by the authors for the first time. They carry no special legitimacy or significance beyond what the authors claim for them. And since the results of the AGPC analysis differ significantly across the three categories, the possibility that these categories are (at best) artefacts or (at worst) targets for data manipulation must be considered.
At first glance, the “ASD-implicated” gene category seems quite reasonable. This set of “genes” (actually a list of 36 genes and 10 loci) includes known genes for several disorders that have shown high susceptibility to autism: conditions like tuberous sclerosis, neurofibromatosis, and syndromes such as Angelman’s, Rett, Fragile X and Smith-Lemli-Opitz.
On the surface, that all seems reasonable. But upon closer inspection, it’s not at all clear how the researchers made their choices between the genes that made it on to the “ASD-implicated” list and the genes that were merely “ASD candidates.” Upon what basis did they draw the line between these two categories? Indeed, they promote their ASD-implicated list as a group of active genes, yet over 20% of their DNA regions are “loci” rather than identified genes. Within the listed genes, while a number of the ASD implicated genes seem to fit their description, quite a few seem arbitrary. For example, why is the NRXN 1 gene “ASD-implicated” and not merely an ASD-candidate? Perhaps because the AGPC is quite favorably inclined to accept the NRXN1 a “known autism gene” (after all, they were the first to suggest it might be an important CNV), but that hardly puts it in the same category as the gene for Fragile X syndrome. And at least one other group found no support for this gene as significant in autism.
This is no trivial matter, with 103 “ASD candidates” providing no significant support for a CNV role in autism and 46 “ASD-implicated” genes and loci providing modest support, the question of whether there is any valid distinction between these two groups is critical. If the AGPC team had analyzed all 149 “genes” as a single group, would there have been a significant finding? I suspect not. Would the subset of the 46 ASD-implicated genes that drove the significance of the result increase the credibility of the group as a whole if identified by name? One wonders. Indeed, the simple assertion that these categories carry biological significance raises questions over how and whether this framework was qualified and tested. Even more troubling, without rigorous criteria for selection of these categories (none are provided in the 74 page supplement) one wonders whether the three lists were created before or after the analysis began. Needless to say, any ex post categorization creates huge potential for bias and even for abuse, with the investigators “tuning” the respective lists to cull out a subset around which they can conjure up a favorable finding. Did the AGPC team hold themselves to a rigorous standard of ex ante hypothesis formulation and empirical testing? One would certainly hope so, but no insight into how this design work was conceived or carried out is described in the methods.
Association is not causation
There is also a basic logic problem in the entire autism-CNV literature that bears emphasizing. It’s one thing to find evidence of differences in CNV rates, either inherited or de novo, in a sample of autistic subjects and quite another thing to draw extravagant conclusions. More specifically, if an exhaustive data mining exercise like the AGPC project can succeed in extracting a faint signal of significance from the vast swath of base pairs that make up the human genome, it’s one thing to report that there might be evidence of a small excess in mutation rates. But to suggest that any constellation of these mutations actually cause autism is a conceptual leap that’s actually quite risky.
As the watchdogs of scientific rigor so frequently remind us, association is not the same as causation.
What does that mean for the interpretation of autism-CNV research? The one glaringly obvious possibility that none of the genetic researchers can bring themselves to admit is that their painstaking genetic findings could simply be an effect and not a cause. Indeed, why wouldn’t it be the case that a population of pregnant women and infants that is under higher environmental stress than a control population would have a higher rate of DNA damage? Isn’t it more likely that the environmental exposures that have provoked the autism epidemic are also mutagenic in their own right?
Mercury, for example, is a well-know mutagen. Many other toxins have similarly genotoxic effects. So instead of assuming a straight line of causality from the genome to autism based on a meandering pattern of findings from the CNV analyses of the AGPC and others, there’s a far more parsimonious interpretation. Any kind of elevated CNV rate in autism (to the extent that any such pattern endures, more on this below) could simply be a benign outcome of causal environmental factors.
And all of this expensive number-crunching could amount to much ado about nothing.
Fire and fall back: gene hunters in a strategic retreat
With methods and designs as creative as these, it shouldn’t surprise anyone that there has been a veritable parade of autism gene studies that have not only failed to reinforce one another, they have consistently repudiated the findings and (often grandiose) theories of previous investigations. The CNV parade has rolled out in just a few short years. In 2007, a group from Cold Spring Harbor (1) reported that de novo CNVs appeared in 10% of their ASD cases and only 2% in controls. Soon after, a group called The Autism Consortium (not the AGPC), based at Harvard, looked for support for a higher rate of de novo mutations in autism (4). Although they reported a “hot spot” of genetic instability on chromosome 16, they found no evidence of the pervasive pattern the Cold Spring Harbor group proposed. Meanwhile, and publishing before the Harvard group, the first AGPC report jumped on the CNV bandwagon as well and pointed loudly to a de novo deletion on the NRXN1 gene (3). Shortly thereafter, however, the Harvard group found no evidence for a significant rate of de novo NRXN1 deletion in autism. In addition, yet another 2008 analysis found that NRXN1 deletions might not be such a big deal after all; controls could have the NRXN1 deletion too--so it wasn’t really causal (6)--and although some affected families had the same deletion (this time it wasn’t de novo but rather in the parents), the inheritance pattern showed that the NRXN1 deletion did little to explain the autism risk (6). Not long after these two 2008 publications poked holes in the 2007 CNV claims, a 2009 study from The Children’s Hospital of Philadelphia (CHOP) threw out the Harvard group’s 2008 “breakthrough,” finding no evidence for any excess of de novo mutations in autism and no “hot spot of genetic instability” on 16p11.2. Instead, the CHOP group turned the focus back to inherited CNVs, proposing several new candidate regions (7). Unfortunately for CHOP, the two most recent analyses (8,9), have rejected their most prominent proposals.
And the beat goes on with the new AGPC report. You’d never know it from the publicity blitz, but the latest AGPC study was bad news for almost all of its predecessors’ CNV findings as it failed to replicate almost all of their most prominent predictions. For one, although the AGPC paper stretched to find a high rate of inherited CNVs, they found no evidence for a higher rate of de novo mutations in autism. They also found no support for elevated CNV rates on a long list of “ASD candidate” genes, regions that include many of the CHOP areas, not to mention others from the list of 103 genes in these regions that have been raised in previous studies. And the faint signals they so aggressively promoted in their public relations program have yet to face a major replication test.
If the past provides any lesson, the latest AGPC claims will likely prove ephemeral, and turn into yet another casualty in a long parade of failed autism gene studies. As the table below demonstrates graphically, the specific pattern of proposing CNV abnormalities in autism (the green shaded areas) only to find that the evidence contradicts the proposed abnormality (the red shaded areas) as new data comes in, has repeated itself multiple times.
(Click table to enlarge.)
So instead of being a startling new breakthrough in autism science, the latest AGPC report is actually just the opposite: yet anotherGroundhog Day repetition of the futility of autism gene science unaccompanied by any acknowledgement of that sad reality. Instead, the strategic retreat is masked by a sharp volley of new claims, with the gene science enterprise firing as it falls back.
To make this point even more clear than I have so far, it’s worth elaborating a bit on how a few specific theories of CNV causation in autism have been forced into retreat. Four of them (most notably represented by the four blue boxes in the table above) make this case most clearly.
1. The unified genetic theory. The 2007 report from the Cold Spring Harbor (CSH) group that first focused attention on CNVs in autism brought with it both an ambitious theory and strong predictions. Unlike most previous genetic work in autism that focused on inheritance, the CSH group brought forth a new and largely environmental theory, calling it a “unified theory for sporadic and inherited autism.” Based on provocative results from a small sample, they argued for a specific pattern of genetic mutations in autism: that rare CNVs occurred at a higher overall rate in autism families (implicitly, this rate reflected environmental pressure and explained the increased rate of autism); that families with multiple affected children (called multiplex families) harbored CNVs that first emerged in the parental germ line and that these parental mutations were passed on, largely through genetically protected females (this was a new variant of mother blaming, the “mutant mom” theory) to multiple male children; and that families with a single affected child (or simplex families) would show an especially high rate of de novo mutations, not present in the parents. According to some private accounts, this bold new theory was all the rage for a short while. It offered a creative new synthesis of genetic transmission patterns and environmental causation (via mutagenesis) that both explained old facts and was sustained by new data in a pilot analysis (2).
1. The unified genetic theory. The 2007 report from the Cold Spring Harbor (CSH) group that first focused attention on CNVs in autism brought with it both an ambitious theory and strong predictions. Unlike most previous genetic work in autism that focused on inheritance, the CSH group brought forth a new and largely environmental theory, calling it a “unified theory for sporadic and inherited autism.” Based on provocative results from a small sample, they argued for a specific pattern of genetic mutations in autism: that rare CNVs occurred at a higher overall rate in autism families (implicitly, this rate reflected environmental pressure and explained the increased rate of autism); that families with multiple affected children (called multiplex families) harbored CNVs that first emerged in the parental germ line and that these parental mutations were passed on, largely through genetically protected females (this was a new variant of mother blaming, the “mutant mom” theory) to multiple male children; and that families with a single affected child (or simplex families) would show an especially high rate of de novo mutations, not present in the parents. According to some private accounts, this bold new theory was all the rage for a short while. It offered a creative new synthesis of genetic transmission patterns and environmental causation (via mutagenesis) that both explained old facts and was sustained by new data in a pilot analysis (2).
Unfortunately for the CSH group, however, their bold predictions failed to hold up under close inspection with larger samples. Relatively few subsequent analyses have examined the rate of de novo mutation in a control group, but the one study that did found no support for a higher mutation rate in autism families (4). One study found a higher de novo mutation rate in simplex families (6), but the new AGPC analysis showed the two rates to be virtually identical (9). More importantly, virtually none of the later evidence supported the unified model of autism gene acquisition or transmission, either from “mutant moms” to multiple affected children or de novo in simplexfamilies. As one study conceded, their results were “consistent with the idea that true risk loci show…imperfect segregation with disease, a reality that will complicate gene finding efforts.” (8) So much for mutant moms.
2. The hot spot theory. Following the fashion of seeking de novo CNVs, the Harvard-based Autism Consortium published their 2008 study suggesting that despite a lack of general de novo risk, as the Cold Spring Harbor theory went, there were specific areas of “genetic instability” on the genome that could be tied to autism risk. Like Cold Spring Harbor, the Harvard group’s theory was consistent with environmental factors playing a role in producing CNVs. Elaborating on this idea, an editorial accompanying the consortium’s paper in The New England Journal of Medicine spun a new grandiose theory of autism based on faint signals from Harvard’s CNV study: that “from an evolutionary standpoint, autism may be a relatively ‘young’ disease’”; that the de novo mutations on chromosome 16p11.2 might represent “a hot spot of genetic instability” in autism; and that these hot spots of de novo CNVs in “rapidly evolving gene families” might “highlight a different paradigm for the genetic basis of autism.” The editorial writers went on to theorize that “each new objective finding expands the number of forms, or ‘autisms,’ like layers of an onion,” making them among the first autism geneticist to blame the failure of their work on the semantics of the disease’s describers rather than the shortcomings of their own theory.
Beyond the obvious extravagance of this leap (that autism is not the singular disorder first observed by Leo Kanner, rather it is thousands of different ones) the hot spot theory failed at a more basic level. Its own purported onion layer, evidence of deletions on 16p11.2, was not a stable finding. The Harvard group’s deletion evidence depended on the idea that the presence of 13 deletions in their cases was significantly higher than the 5 deletions they found in controls. But just a year after their widely publicized finding of a hot spot on 16p11.2, the CHOP group found “the CNV frequency in the control subjects at this locus was…comparable to the cases.” (7) So much for the hot spot.
3. Specific pathways rather than genes. The failure of the de novo mutation theory led the CNV theorists to a new retreat. Out went both the emphasis on genetic instability (inherited CNVs were back in the game) and on specific DNA regions, hot spots or otherwise. In came a broader search for defective biological pathways. In this new model, it didn’t matter whether specific genes could be shown consistently to have mutated in autism families. Instead the goal was to find defects in any numbers of genes that coded instead forbiologically connected proteins. And since the genes no longer held the key, but instead gene clusters, a whole new gene hunting strategy could be deployed. Moving from long lists of genes to short list of pathways got the rhetorical juices flowing. One 2008 paper from the Harvard group theorized that the cause of autism could be traced to metabolic pathways surrounding “the autistic neuron” (10) and proposed that such pathways could be tied to autism by investigating targets such as “synaptic protein synthesis.” Following this idea, the CHOP group made a splash early last year with a report suggesting that both “ubiquitin degradation” and “neuronal gene” pathways (7) were enriched with CNVs
It wasn’t long, however, before the first of these findings bit the dust. Late last year, yet another consortium joined the quest and reported that “no evidence for…overrepresentation was found for genes in the ubiquitin degradation pathway and neither term was highlighted as overrepresented among [deletions] or [duplications].” (8) So much for the short list of pathways.
4. Pathways here, pathways there, defective pathways are everywhere. Not surprisingly, as neither specific genes, nor hot spots, nor specific pathways have held up under scrutiny, the next move of the autism gene hunters has been to retreat to the last refuge they can find, any pathway that offers the hope for a clustering of autism defects. Seen in the context of these successive retreats, it becomes clearer just how the latest AGPC paper fits in. Offering no theory of environmental interaction with the genome while also gliding by a remarkable repudiation of decades of autism gene candidates, they proposed a new retreat. The evidence for their substitute theory? An undifferentiated collection of mutations only in “genic regions” that are traceable only to an unspecified collection of “gene sets.” This is an explanation for a disorder so specific that Leo Kanner memorably described its victims as “children whose condition differs so markedly and uniquely from anything reported so far, that each case merits…a detailed consideration of its fascinating peculiarities”?
In the absurdity of the latest retreat, perhaps we might dare to ask whether it’s time. Could this new paper--a grand mush of random, rare, multiple inherited gene mutations that at best can explain only a tiny fraction of autism cases—be the last stand of the rare CNV theory?
There’s a bigger game being played
As we consider this question, it’s important to remember, however, that in all the storm and fury surrounding the latest AGPC findings there is more at stake than the mere collapse of autism gene science. The larger failure of genetic research goes well beyond autism and has far more serious consequences for the greatest hope of 20th century biology. Indeed, one reason autism gene studies get so much attention is that autism was always supposed to be an early victory, one of the easiest targets in the broader search for the genetic roots of human disease. If genetic research can’t prove its value in autism, then the prospects for the larger enterprise are bleak indeed. The New York Times, hardly a harsh critic of the great autism gene hunt over the years, wrote about this problem on June 12th, in an article titled, “A Decade Later, Genetic Map Yields Few New Cures.” (see HERE)
“For biologists, the genome has yielded one insightful surprise after another. But the primary goal of the $3 billion Human Genome Project — to ferret out the genetic roots of common diseases like cancer and Alzheimer’s and then generate treatments — remains largely elusive. Indeed, after 10 years of effort, geneticists are almost back to square one in knowing where to look for the roots of common disease.”
This failure has been particularly acute in the effort to find common (mostly inherited) disease genes, including any involved in autism (for a related discussion, see HERE).
“With the catalog [of common variants] in hand, the second stage was to see if any of the variants were more common in the patients with a given disease than in healthy people. These studies required large numbers of patients and cost several million dollars apiece. Nearly 400 of them had been completed by 2009. The upshot is that hundreds of common genetic variants have now been statistically linked with various diseases. But with most diseases, the common variants have turned out to explain just a fraction of the genetic risk.“
That failure has forced a retreat that goes beyond those we’re seeing in autism, but one that helps explain the recent infatuation with autism CNV research; it’s a retreat from the specification of the catalog of common genes to a search for rare mutations, or CNVs. According to The New York Times report, “It now seems more likely that each common disease is mostly caused by large numbers of rare variants, ones too rare to have been cataloged.”
Seen from a larger perspective, then, the AGPC celebration was all about the struggle with this looming specter of failure: an attempt to avoid the shame of yet another defeat, to loudly claim victory in what is in truth another losing skirmish in a war the gene hunters are losing badly. On the autism front, once thought easily winnable, the larger genetics enterprise is now scattering backward after absorbing multiple blows to their theory and is beating several different paths of retreat all at once: a retreat from linkage studies that aimed to find common inherited genes; a retreat from grand, unified theories of de novo mutation; a retreat from hot spots of genetic instability; and (latest) a retreat from short lists of candidate pathways to an undifferentiated mush of 75 gene sets representing a multitude of only dimly connected pathways.
Despite this historic sequence of failure, the AGPC group understands their need to defend the millions of dollars of investment in the great autism gene hunt, and so again they have positioned this latest publication as a celebration when an incoherent mush of findings offered no support whatever for decades of prior work, “explained” only 2% of autism with a questionable clustering of (at least some) known risk genes and stretched that claim to 3.3% with a brand new list that had nothing to do with autism.
Then, in an astonishing bit of hubris, the AGPC went on to double down on their bets, pitching parents directly for more money and more community support. In an Autism Speaks blog post written shortly after the paper’s publication, AGPC member Stanley Nelson made a plea to the autism parent community (see HERE) to invest even more time and more money in gene hunting research. His idea? That chasing down all of these faint signals would require sample populations that were many times larger than those assembled for past work, populations that already number in the thousands
You have to hand it to these guys; they sure know how to sell a losing war.
Keep in mind that this long, slow collapse of the great autism gene hunt is no cause for celebration in its own right. This tragic waste of time and money is both surprising and unfortunate. It shouldn’t have been this hard to gain genetic insight into the biology of autism. Everyone knows there is familial clustering in autism. Everyone concedes that there ought to be genetic susceptibility factors. And moving faster to learn the lessons of autism’s underlying biology is something all of us should be rooting for. Yet, autism is teaching us another lesson instead: about the futility of looking to our genes for explanations of the cause of human disease. That’s a painful lesson for many.
But when it comes to autism, it should also be clear that the massive hype surrounding the new AGPC study was out of all proportion to the importance of its scientific findings, not to mention any benefits it offered autism families. The AGPC report found almost nothing yet claimed almost everything. Worse, they succeeded in attracting millions of dollars on speculative work that has generated zero value to date and in the process crowded out more promising avenues of research. Optimistic headlines aside, no predictive genetic tests are in sight and nor is there much prospect for new treatment targets. Even if there were a faint hope of new drug target targets tied to the putative list of target gene clusters, any drug development that might stem from this work lies years, if not decades, in the future. For the foreseeable future, there will be no magic pill for autism. The best one might hope for is a ruthless plan for prevention via genetic counseling, i.e. abortion.
It’s time. Time to put an end to the great autism gene hunt. Time to turn resources over to the investigation of environmental risk factors. Time to turn from delusion to rationality. Time, in the words of Bernard Rimland, to defeat autism.
Time for our children (and an exploding number of young adults) is growing short. Let’s not waste any more of it.
References:
1: Sebat J, Lakshmi B, Malhotra D, et al. Strong association of de novo copy number mutations with autism. Science. 2007;316(5823):445-9.
1: Sebat J, Lakshmi B, Malhotra D, et al. Strong association of de novo copy number mutations with autism. Science. 2007;316(5823):445-9.
2: Zhao X, Leotta A, Kustanovich V, et al. A unified genetic theory for sporadic and inherited autism. Proc Natl Acad Sci U S A. 2007;104(31):12831-6.
3. Szatmari P, Paterson AD, Zwaigenbaum L, et al. (Autism Genome Project Consortium) Mapping autism risk loci using genetic linkage and chromosomal rearrangements. Nat Genet. 2007;39(3):319-28.
4. Weiss LA, Shen Y, Korn JM, et al. (Autism Consortium) Association between microdeletion and microduplication at 16p11.2 and autism. N Engl J Med. 2008;358(7):667-75.
5. Eichler EE, Zimmerman AW. A hot spot of genetic instability in autism. N Engl J Med. 2008;358(7):737-9.
6. Marshall CR, Noor A, Vincent JB, et al. Structural variation of chromosomes in autism spectrum disorder. Am J Hum Genet. 2008;82(2):477-88.
7: Glessner JT, Wang K, Cai G, et al. Autism genome-wide copy number variation reveals ubiquitin and neuronal genes. Nature. 2009;459(7246):569-73.
8. Bucan M, Abrahams BS, Wang K, et al. Genome-wide analyses of exonic copy number variants in a family-based study point to novel autism susceptibility genes. PLoS Genet. 2009;5(6):e1000536.
9. Pinto D, Pagnamenta AT, Klei L, et al. (Autism Genome Project Consortium) Functional impact of global rare copy number variation in autism spectrum disorders. Nature. 2010 Jun 9. [Epub ahead of print]
10. Kelleher RJ 3rd, Bear MF. The autistic neuron: troubled translation? Cell. 2008;135(3):401-6
