Only Some of These Really Bother Me

(A version of this post has also appeared on my Tumblr)

Article header from io9.com: "10 Scientific Ideas That Scientists Wish You Would Stop Misusing"

There's an article on io9.com listing ten words/concepts from various fields of science that are commonly misused by laypeople, so I had to look at it and see if any of my pet peeves made it onto the list.

There were a few:

3. Quantum Uncertainty and Quantum Weirdness
[Astrophysicist Dave] Goldberg adds that there's another idea that's been misinterpreted even more perniciously than "theory." It's when people appropriate concepts from physics for new agey or spiritual purposes:

This misconception is an exploitation of quantum mechanics by a certain breed spiritualists and self-helpers, and epitomized by the abomination, [the movie] What the Bleep Do We Know? Quantum mechanics, famously, has measurement at its core. An observer measuring position or momentum or energy causes the "wavefunction to collapse," non-deterministically. (Indeed, I did one of my first columns on "How smart do you need to collapse a wavefunction?") But just because the universe isn't deterministic doesn't mean that you are the one controlling it. It is remarkable (and frankly, alarming) the degree to which quantum uncertainty and quantum weirdness get inextricably bound up in certain circles with the idea of a soul, or humans controlling the universe, or some other pseudoscience. In the end, we are made of quantum particles (protons, neutrons, electrons) and are part of the quantum universe. That is cool, of course, but only in the sense that all of physics is cool. 

4. Learned vs. Innate
Evolutionary biologist Marlene Zuk says:

One of my favorite [misuses] is the idea of behavior being "learned vs. innate" or any of the other nature-nurture versions of this. The first question I often get when I talk about a behavior is whether it's "genetic" or not, which is a misunderstanding because ALL traits, all the time, are the result of input from the genes and input from the environment. Only a difference between traits, and not the trait itself, can be genetic or learned — like if you have identical twins reared in different environments and they do something different (like speak different languages), then that difference is learned. But speaking French or Italian or whatever isn't totally learned in and of itself, because obviously one has to have a certain genetic background to be able to speak at all.

....

6. Gene

[Synthetic biologist Terry] Johnson has an even bigger concern with how the word gene gets used, however: 

It took 25 scientists two contentious days to come up with: "a locatable region of genomic sequence, corresponding to a unit of inheritance, which is associated with regulatory regions, transcribed regions and/or other functional sequence regions." Meaning that a gene is a discrete bit of DNA that we can point to and say, "that makes something, or regulates the making of something". The definition has a lot of wiggle room by design; it wasn't long ago that we thought that most of our DNA didn't do anything at all. We called it "junk DNA", but we're discovering that much of that junk has purposes that weren't immediately obvious. 

Typically "gene" is misused most when followed by "for". There's two problems with this. We all have genes for hemoglobin, but we don't all have sickle cell anemia. Different people have different versions of the hemoglobin gene, called alleles. There are hemoglobin alleles which are associated with sickle cell diseases, and others that aren't. So, a gene refers to a family of alleles, and only a few members of that family, if any, are associated with diseases or disorders. The gene isn't bad - trust me, you won't live long without hemoglobin - though the particular version of hemoglobin that you have could be problematic. 

I worry most about the popularization of the idea that when a genetic variation is correlated with something, it is the "gene for" that something. The language suggests that "this gene causes heart disease", when the reality is usually, "people that have this allele seem to have a slightly higher incidence of heart disease, but we don't know why, and maybe there are compensating advantages to this allele that we didn't notice because we weren't looking for them".

Those were the ones that resonated with me the most; others were only minor peeves or didn't actually bother me at all.

Misused Word #1, "Proof," was only a minor annoyance for me in that I'm almost never talking about mathematical proofs, and even if I were the sort of person who does use them routinely, it still seems to me like most things people talking about "proving" colloquially are impossible to express in mathematical terms.

It just seems to me like there wouldn't be very many circumstances in which mixing up the technical and colloquial meanings of "proof" would be an issue that would even arise.

(I have found that the most annoying sources of confusion in scientist/layperson conversations about proof have to do with standards of evidence, or also degrees of uncertainty. You can be more unsure of one thing than you are of another, even if you're not 100% certain about the thing you are more sure of.)

Similarly, "theory" also doesn't annoy me that much because I don't usually have much trouble adjusting to different usages of words in different contexts.

I can see how it would get really old having to explain the technical meaning of "theory" over and over again, though.

It seems like those are more about the meaning of specific words than they are about whole networks of ideas, so they are easier for me to adapt to when they surprise me in conversation.

The ones discussed in the quoted text above, though? Misuse of ideas derived from quantum mechanics, misinterpretations of evolution and natural selection, or the idea that genes are "for" specific things? Those come with so many other ideas connected to them, so many wrong things tacitly accepted as premises, that I feel like I need a ball of yarn to slowly pick my way back to the start of the conceptual maze.

Autism-Related Gene Spotlight: MECP2

Where is it?
Near the very end of the X chromosome, at Xq28.
Here is a picture of its position relative to some other genes at that part of the X chromosome:
You can see that it's not the last gene on there, and there are quite a few known and potential genes following it, but it's really, really close to the end. That picture I just posted? With MECP2 appearing at the far left? That's the very end of a 24-page image. So, based on that I feel comfortable calling MECP2 one of the last genes on the X chromosome.

What does it do?It encodes a protein, MeCP2, that can bind to methylated DNA (and also to a variety of other transcription-repressing proteins) and whose function is to repress transcription of its target genes. (More recent research has also found that it can also serve as a transcriptional activator). It has a lot of target genes, and their functions vary widely; many of them are other transcription factors, and many are involved in cell-cell signaling, or in signal transduction within the cell. Overall, transcription and neurotransmission seem to be the physiological processes that the majority of MeCP2 target genes are involved with, though it is also important for nerve and muscle cell growth (and thus, needs to be expressed in different amounts at different times during development). It is highly expressed in nerve cells. It's also been found to have other functions, like RNA splicing, chromatin remodeling and DNA methylation.

What mutant versions of this gene have been discovered?
(Here's a very rough impression of where some of the more common mutations (and some less-common ones that I talk about in the next section) associated with Rett syndrome fall on a map of MECP2 coding regions. Mutations that only change an amino acid are outlined in different shades of red-orange; mutations that produce a truncated version of the MeCP2 protein are outlined in black, and indicated on the map with little stop signs. Image not drawn to scale)

This 1999 article in Nature Genetics (full text here) describes a genetic analysis of 29 girls with Rett syndrome* (8 of whom had a family history of the condition), which found seven point mutations (changes in a single nucleotide) and one case where an extra nucleotide (thymine) was inserted into the gene, which threw off the "reading" of everything that came after, since protein synthesis depends on grouping the nucleotides into threes, and stringing together the amino acids corresponding to each sequence of three nucleotides, or "codons". Changing one nucleotide to another will therefore change one amino acid in the resulting protein, while adding or subtracting a nucleotide will change every amino acid that follows. (Such "frameshift" mutations are much more likely than point mutations to result in a nonfunctional protein).

They were: a substitution of cytosine for thymine at nucleotide #538; substitutions of thymine for cytosine at nucleotides #390, #471, #547, #656, #837, and #1307; and the aforementioned insertion of (an extra) thymine between nucleotides #694 and #695.

Another genetic analysis described in a 2000 article in Human Molecular Genetics found 17 different mutations in 46 girls with Rett syndrome; these mutations included substitutions of thymine for cytosine at nucleotides #473, #502, #763, #808, #880, and #916; substitutions of guanine for cytosine at nucleotides #905 and #1038; a substitution of thymine for adenine at nucleotide #592; a substitution of cytosine for adenine at nucleotide #1461; a substitution of adenine for guanine at nucleotide #317; and a ten-nucleotide deletion starting at nucleotide #1158. Most of these mutations were in exon 3, though there were a few in exons 2 and 4 as well.

A 2004 analysis of DNA samples from 56 French women and girls with Rett syndrome found five frameshift mutations: a deletion of nucleotide #345, in exon 3; a deletion 202 nucleotides long, starting at position #895; another deletion 53 nucleotides long starting at position #1124; a deletion of 8 nucleotides and an insertion of 18 nucleotides starting at position #989; and an insertion of an AG dinucleotide after nucleotide #996. All of these last four were in exon 4.

An article from this year describes a 41-base deletion in a Korean girl with Rett syndrome; the deleted region started at nucleotide #1152, in exon 4.

Another article from this year found a substitution of thymine for cytosine at nucleotide #535 in a Tunisian girl with Rett syndrome.

This article (full text here) describes 17 mutations: a substitution of thymine for guanine at nucleotide #298; a substitution of adenine for guanine at nucleotide #398; a substitution of guanine for adenine at nucleotide #914; a substitution of thymine for cytosine at nucleotide #730; an insertion of (an extra) guanine after nucleotide #704; an insertion of cytosine after nucleotide #747; and multiple deletions, most of which had starting points between nucleotides 1,000 and 1,200, and all but one of which were deletions of multiple nucleotides. There was also a sequence of 137 nucleotides, starting at position #1169, that was repeated.

This 2009 genomic analysis of 74 people with Rett syndrome in New Zealand turned up four new mutations, including a fairly large deletion (1,596 nucleotides) that encompassed both exons 3 and 4.

There are a lot more --- the International Rett Syndrome Foundation's database of mutations associated with Rett syndrome (RettBASE) lists 4,225 different mutations. Not all of them are in MECP2, but a large majority of them are.

Mutations in MECP2 can also be associated with conditions other than Rett syndrome: this article describes mutations found in five children with Angelman syndrome. Two of them had deletions in exon 4, one had a two-nucleotide deletion in exon 3, and the others had single-base substitutions.

How do these mutations affect protein function?
The MeCP2 protein has two regions (called domains) that are crucial to its function in the cell: the methyl-DNA binding domain (MBD), which allows it to bind to methylated cytosines, and the transcription repression domain (TRD), which binds to other enzymes that condense chromosomal DNA and make it impossible for the enzymes reponsible for transcription to bind to it. MeCP2's role in transcription repression seems to be to bring the enzymes that do the actual repressing to its target sequences of DNA, rather than to block transcription itself.
(Image of the structure of the MeCP2 methylDNA-binding domain, showing the amino acids affected by some of the more common mutations)

The MECP2 gene has four exons, of which three contain sequences encoding these domains: Exon 2 encodes most of the DNA-binding domain, with some of it spilling over into exon 3, and parts of exons 3 and 4 encode the transcription repressor domain. So, depending on where it occurs in the gene, a mutation might disrupt either the MeCP2 protein's DNA-binding capacity, or its ability to bind to those other, transcription-repressing enzymes.

Most of the mutations associated with Rett syndrome (or other conditions mentioned in the above section) change the structure of one of those domains in such a way as to weaken, or completely destroy, its ability to bind to whatever it needs to bind to. This article describes the effect on DNA binding ability of several known mutations (including a few of the most common ones) that alter the amino-acid sequence of the MBD. The mutation with the greatest effect on MeCP2's DNA-binding ability, p.R111G, swaps out a positively-charged amino acid on the long, flexible loop within the MBD for a nonpolar one; since that loop normally lies close to the sugar-and-phosphate "backbone" of the DNA (the part of the DNA to which the A's, T's, G's and C's all attach, and which forms the two outer ridges of the double helix), and since that backbone carries a negative charge (from all the phosphate groups), knocking out positively-charged amino acids in this region will disrupt the attraction between the DNA and the methylDNA-binding region of MeCP2.

Another mutation that can cause a sharp decline in DNA-binding ability, which also happens to be one of the most commonly-occurring mutations in people with Rett syndrome, is p.R133C, which also replaces a positively-charged amino acid with a nonpolar one. This one occurs in a different part of the MBD than p.R111G does, a "beta sheet" made up of long, flat strings of amino acids laid side by side. One of the short loops connecting two of the component strands has a sequence of five amino acids with hydrophobic side chains that create a "pocket" sequestering the methyl groups attached to the DNA. It may not always lead to loss of function, though; this group of mostly Japanese researchers conducted a similar analysis (full text here) of protein function, comparing some of the most common mutant versions of MeCP2 with its normal, "wild-type" form, and they found that the R133C variant bound to DNA almost as readily as the wild-type MeCP2 did.

Other mutations associated with a near-total loss of DNA-binding ability are p.G114P, which replaces an amino acid in the middle of the long, flexible loop described above with one whose rigidly-structured, bulkier sidechain would greatly restrict the loop's ability to move and re-fold itself to fit into the groove of the DNA helix; p.D121A and p.D121E, which substitute amino acids with, respectively, nonpolar and negatively-charged sidechains for one with a positively-charged sidechain on one of the strands of the beta-sheet comprising another of the MBD's DNA-contacting surfaces; two other fairly common mutations, p.R106W and p.F155S, throw off the protein's overall folding to such an extent that it becomes unstable at body temperature.

Several mutations cause transcription of MECP2 to stop prematurely, leading to the production of an incomplete protein. Depending on where the erroneous "stop" signal occurs, the resulting protein might be missing all or part of its transcription-repressor domain.

Mutations occurring downstream of the transcription-repressor domain have also been associated with problems; this experiment showed that mutant versions of MeCP2 that don't have the long tail following the TRD are less stable than wild-type MeCP2, and tend to break down quickly in the cellular environment.

How common are they?
This article in the European Journal of Human Genetics lists eight MECP2 mutations its authors consider "common," along with each mutation's prevalence among the people with Rett syndrome listed in either the British Isles Rett Survey or the Australian Rett Syndrome Database. Of the 524 cases they looked at, 65 (12.8%) had the mutation p.T158M, which is the substitution of thymine for cytosine at nucleotide #473; 58 (11.1%) had the mutation p.R168X, which is the substitution of thymine for cytosine at nucleotide #502; 44 (8.4%) had the mutation p.R270X, which is the substitution of thymine for cytosine at nucleotide #808; and 42 (8%) had the mutation p.R255X, which is the substitution of thymine for cytosine at nucleotide #763. The other four mutations listed as "common" in this paper --- p.R106W (thymine substituted for cytosine at nucleotide 316), p.R133C (thymine for cytosine at nucleotide 397), p.R294X (thymine for cytosine at nucleotide 880) and p.R306C (thymine for cytosine at nucleotide 916) all account for between 3 and 7 percent of all cases surveyed.

Another article (full text here) also found those eight mutations occurred several times in their sample of 116 people with Rett syndrome; these researchers also found p.T158M to be the most common, present in 12 different people. (The next-most common ones were p.R270X, found in eight people, and p.R255X and p.R106W, each found in seven people). This study also listed three other mutations in its table of "recurring" mutations: a substitution of guanine for cytosine at nucleotide 455 (observed four times), a substitution of thymine for cytosine at nucleotide 965 (observed twice), and a modification of a splice site in exon 4 (an AG sequence becomes GG; this permutation was also observed only twice).

RettBASE also ranks the various mutations by frequency of occurrence: there, too, p.T158M is the most common, with 363 known occurrences and accounting for 8.59% of all mutations identified so far. Most of the mutations (about two-thirds) listed there are unique.

Rett syndrome occurs in between 1:10,000 and 1:22,000 girls, and has only been recorded in 20 boys, ever. (Usually if a boy is born with the kind of mutations that would lead to Rett syndrome in a girl, he dies). So when I say a given mutation is found in, say, 10% of all people with Rett syndrome, that would translate into between 1:100,000 and 1:220,000 for its frequency among all people. So, while some MECP2 mutations might be less rare than others, I'd say they're all rare.
Database entries for this gene: AutDB, Ensembl, Entrez Gene, GeneCards, Genetics Home Reference, WikiGenes

Sources:
Amir RE, Van den Veyver IB, Wan M, Tran CQ, Francke U, & Zoghbi HY (1999). Rett syndrome is caused by mutations in X-linked MECP2, encoding methyl-CpG-binding protein 2. Nature genetics, 23 (2), 185-188 PMID: 10508514

Bienvenu, T. (2000). MECP2 mutations account for most cases of typical forms of Rett syndrome Human Molecular Genetics, 9 (9), 1377-1384 DOI: 10.1093/hmg/9.9.1377

Bienvenu T, Souville I, Poirier K, Aquaviva C, Burglen L, Amiel J, Héron B, Kaminska A, Couvert P, Beldjord C, & Chelly J (2001). Five novel frameshift mutations in exon 3 and 4 of the MECP2 gene identified in Rett patients: Consequences for the molecular diagnosis strategy. Human mutation, 18 (3), 251-252 PMID: 11524737

Díaz de León-Guerrero, S., Pedraza-Alva, G., & Pérez-Martínez, L. (2011). In sickness and in health: the role of methyl-CpG binding protein 2 in the central nervous system European Journal of Neuroscience, 33 (9), 1563-1574 DOI: 10.1111/j.1460-9568.2011.07658.x

Fendri-Kriaa N, Hsairi I, Kifagi C, Ellouze E, Mkaouar-Rebai E, Triki C, Fakhfakh F, & The Tunisian network on mental retardation study (2011). A case of a Tunisian Rett patient with a novel double-mutation of the MECP2 gene. Biochemical and biophysical research communications, 409 (2), 270-274 PMID: 21575601

Free, Andrew, Robert I. D. Wakefield, Brian O. Smith, David T. F. Dryden, Paul N. Barlow, & Adrian P. Bird (2000). DNA Recognition by the Methyl-CpG Binding Domain of MeCP2 Journal of Biological Chemistry, 276 (5), 3353-3360 DOI: 10.1074/jbc.M007224200

Hite, K., Adams, V., & Hansen, J. (2009). Recent advances in MeCP2 structure and function Biochemistry and Cell Biology, 87 (1), 219-227 DOI: 10.1139/o08-115

Hoffbuhr K, Devaney JM, LaFleur B, Sirianni N, Scacheri C, Giron J, Schuette J, Innis J, Marino M, Philippart M, Narayanan V, Umansky R, Kronn D, Hoffman EP, & Naidu S (2001). MeCP2 mutations in children with and without the phenotype of Rett syndrome. Neurology, 56 (11), 1486-1495 PMID: 11402105

Kudo, S., Y. Nomura, M. Segawa, N. Fujita, M. Nakao, C. Schanen, & M. Tamura (2003). Heterogeneity in residual function of MeCP2 carrying missense mutations in the methyl CpG binding domain Journal of Medical Genetics, 40 (7), 487-493 DOI: 10.1136/jmg.40.7.487

Kumar, A., Kamboj, S., Malone, B., Kudo, S., Twiss, J., Czymmek, K., LaSalle, J., & Schanen, N. (2008). Analysis of protein domains and Rett syndrome mutations indicate that multiple regions influence chromatin-binding dynamics of the chromatin-associated protein MECP2 in vivo Journal of Cell Science, 121 (7), 1128-1137 DOI: 10.1242/jcs.016865

Lee EY, Chung HJ, Ki CS, Yoo JH, & Choi JR (2011). A novel mutation in the MECP2 gene in a Korean patient with Rett syndrome. Annals of clinical and laboratory science, 41 (1), 93-96 PMID: 21325263

Raizis AM, Saleem M, MacKay R, & George PM (2009). Spectrum of MECP2 mutations in New Zealand Rett syndrome patients. The New Zealand medical journal, 122 (1296), 21-28 PMID: 19652677

Singh, J., Saxena, A., Christodoulou, J., & Ravine, D. (2008). MECP2 genomic structure and function: insights from ENCODE Nucleic Acids Research, 36 (19), 6035-6047 DOI: 10.1093/nar/gkn591

Yusufzai, Timur M., & Wolffe, Alan P. (2000). Functional consequences of Rett syndrome mutations on human MeCP2 Nucleic Acids Research, 28 (21), 4172-4179 DOI: 10.1093/nar/28.21.4172

Gene Expression in Autistic Brain Tissue

EXECUTIVE SUMMARY: A recent analysis of mRNA extracted from brain tissue samples taken from brains donated to the Autism Tissue Project and the Harvard Brain Bank, with both sources supplying both autistic and neurotypical brains, found two clusters of genes whose expression differs significantly between autistic and NT brain tissues. One of these clusters, whose component genes primarily encode proteins involved in synapse formation and neurotransmission, is expressed less in the autistic brains than it is in the NT ones. The other cluster includes a lot of immune- and inflammation-related genes, and those genes are expressed to a greater degree in autistic brains. This study also unearthed a gene encoding a protein that seems to be involved in splicing the mRNA transcripts of other genes, and the underexpression of this gene in autism may be part of the reason so many of the other genes in the same cluster are also underexpressed.
_______________________________________________
This study, published online at Nature.com last Wednesday, strikes me as an interesting hybrid of two kinds of studies that are frequently used in autism research: gene-expression studies (where researchers compare patterns of gene expression in tissue samples taken from autistic people to those in samples taken from neurotypical people) and neuroanatomical studies (where researchers look at differences in size, structure or activation patterns of various brain structures between autistic and neurotypical subjects).

In this study, the researchers took samples from three different regions of the brain (the prefrontal cortex, superior temporal gyrus and cerebellar vermis), using brains donated to the Autism Tissue Project and the Harvard brain bank. From both of those sources, they ended up with 58 cortex samples (29 from autistic brains; 29 from non-autistic brains) and 21 cerebellar samples (11 autistic; 10 not).

They looked for differences in gene expression between autistic and control tissue samples by measuring the amount of RNA present in each sample corresponding to a given sequence of genomic DNA. (Since the mid-to-late 1990s, a tool has existed to do this at quite a high resolution: the DNA microarray. This is a glass or silicon chip covered in tiny wells where a short sequence of DNA is anchored --- in this case, the DNA probes are short, synthesized 50-base snippets made to match specific parts of each gene contained in the National Center for Biotechnology Information's RefSeq database --- to which your sample DNA or RNA will bind if it contains a complementary sequence).

(What a microarray looks like)

This type of experimental design doesn't really allow direct comparison between RNA extracts from different tissue samples --- instead, each sample (from the same region of the brain of either an autistic or neurotypical donor, or from different regions within the same brain) hybridizes (i.e., forms a new, DNA-RNA "hybrid" helix when heated) to the DNA probes on its own separate array. Researchers then compare the intensity of the signal created by each well across the two arrays --- the brighter the color, the more (fluorescent-dye-treated) sample RNA is present on the chip.

(DNA-RNA hybridization)

In this experiment, the researchers found 444 genes that differed significantly between the autistic and control cortex* samples in how much mRNA (the RNA created when DNA is transcribed; used as a template for protein synthesis) there was in a given tissue extract. They decided to concentrate on the 200 most differentially expressed genes for their (more detailed) expression analysis, which included a look into "co-expression networks" of genes whose expression seems to be regulated via the same pathways. They replicated their results by doing a similar microarray experiment on tissue samples from a smaller group of donated brains (nine from autistic donors, five from neurotypical donors), this time taking samples from a different region of the cerebral cortex than either of the cortical regions assayed in the initial experiment.
Figure 1C, in Voineagu et al, 2011 --- scatter plot showing genes found to be up- (red) and down-regulated (green) in both the initial and replication data sets. The bluish lines drawn through each cluster of dots reflects the cutoff for significance; most of the regulation changes are significant, but some aren't. Both axes represent a logarithmic measure of the change from baseline for either data set (Data Set 1, or the initial data, is on the x-axis; Data Set 2, or the replication data, is on the y-axis).
They got additional confirmation of their gene-expression data by using an alternative measure of how much of a given mRNA sequence was present in each tissue extract: for each gene that the microarray experiments identified as being differentially expressed in autistic and non-autistic brain tissues, they made a DNA copy of its array-bound mRNAs, and then amplified those bits of DNA using a process called RT-PCR (real-time polymerase chain reaction). That process uses bacterial enzymes to make huge numbers of copies of a given sequence of DNA, such that the amount of DNA is large enough to be easily quantified. Those amounts could be compared between groups, and thus confirm (or fail to confirm) differences in mRNA production predicted by the microarray experiment.

This figure shows how much more or less mRNA there was from eleven genes whose differential expression levels were validated using PCR:

(Supplementary Figure 2B, in Voineagu et al., 2011 --- top bar graph, in red, shows positive changes in expression of five genes in the tissue samples taken from autistic donors; bottom graph, with green bars, shows the average reduction in expression of six other genes. On both graphs, the numbers on the y-axis represent how many times as much of one kind of mRNA was found in the autistic sample; you can see that the up-regulating produced a more dramatic change --- ten- and twenty-fold, for all but one of the five genes --- than down-regulating, which produced, respectively, one-half, one-fifth, one-eighth, one-third and one-fourth as much mRNA as the control sample).

They found two co-expression modules (networks of genes) whose expression varied in relation to whether the sample came from an autistic or non-autistic donor, and not in relation to any of the other variables they took into account (like age, sex, cause of death, medication history, whether the person also had seizures, and family history of mental illness): M12 and M16.

Here is their drawing of M12, and the relationships between its component genes:
(One of the genes in the middle of this diagram, CNTNAP1, is a close relative of a gene that other genetic studies have tied to autism --- and that I have described on this blog --- CNTNAP2).
... and here is their drawing of M16:
Relative to the samples from neurotypical donors, the brain tissue samples from autistic donors had more mRNA transcripts of genes in M16, and fewer transcripts of genes from M12.

In each of these modules, genes for certain types of proteins predominated: for M12, these are proteins involved in synapse formation, neurotransmission, vesicular transport (importing objects into the cell, or exporting objects from it); while M16 included lots of genes for immune and inflammatory proteins.

One of the major genes in M12, A2BP1, is a splicing regulator. Alternative splicing is one of the ways the cell can make different kinds of proteins from the same mRNA; the mRNA will contain characteristic sequences, called splice sites, where splicing enzymes can bind to it, cut it and put it back together, minus the regions bordered by splice sites.

Like some other instances I've mentioned of proteins playing a role in gene expression themselves being expressed differently in autism, this down-regulation of A2BP1 could have important ramifications for the genes whose transcripts A2BP1 is involved in splicing. The authors of this study thought it would be a good idea to look for A2BP1 splice sites in the RNA samples from those specimens within the autism group with especially low levels of A2BP1 mRNA; to do this, they sequenced all the mRNA from three samples with relatively little A2BP1 mRNA, and also from three control samples with normal A2BP1 expression. They found 212 potential splice sites using this method, which they validated by using RT-PCR (again) to compare relative amounts of various alternatively-spliced mRNAs in autistic and control tissue samples --- first in the same three samples that were sequenced, and then in three other samples from the autism group, which also had low A2BP1 expression. Using this method, they confirmed that the vast majority (85%) of the expected splicing changes were really there in all of the low-A2BP1 samples.

The genes whose alternative splicing depended on A2BP1, and thus whose alternate forms were underexpressed in the low-A2BP1 RNA samples, included a lot of the same genes as the M12 co-expression module. So it looks like, besides finding out that M12 is collectively underexpressed in autism, these researchers have also found at least one of the mechanisms behind this underexpression.
*The cerebellar samples differed significantly in the expression of only two genes, so those data were not included in further analysis.

Irina Voineagu, Xinchen Wang, Patrick Johnston, Jennifer K. Lowe, Yuan Tian, Steve Horvath, Jonathan Mill, Rita M. Cantor, Benjamin J. Blencowe, & Daniel H. Geschwind (2011). Transcriptomic analysis of autistic brain reveals convergent molecular pathology Nature (25 May) : 10.1038/nature10110

Making the Genes Fit: Genetic Explanations for Autism and Their Political Implications

EXECUTIVE SUMMARY: Political-science professor Kristin Bumiller has written another long article on the politics of autism; while her earlier article focused on the neurodiversity movement, this article is mostly about mainstream autism advocacy in the vein of Autism Speaks. In it, she argues that mainstream discourse about autism assumes that autism is a genetic condition, and that this assumption is insufficiently backed by evidence. She spends most of the article detailing the political implications of this assumption, which are 1) funnelling most activism on the part of families of autistic people into relatively narrow channels of corporate-philanthropic fundraising for biomedical autism-research initiatives, rather than calling for broader social changes that might benefit autistic people; and 2) making disability an individual, rather than a social, issue.

She introduces several concepts over the course of developing those themes: "geneticization" --- a process by which the preferred explanation for sickness and disability is that some people are genetically susceptible to certain illnesses; "genetic citizenship" --- an ethic of individual responsibility for health, and for knowledge of one's genetic predispositions; and "life optimization" --- a strategy for making the most of one's (or one's child's) life chances given a certain set of genetic predispositions. She makes the case that the latter two of these things are logical responses to a genetic understanding of disease and disability in a "neoliberal welfare state" like the U.S., but also that they are inherently coercive and inegalitarian.
____________________________________________________

Looking through the online archive of the women's-studies journal Signs, I found another article on autism by the political scientist Kristin Bumiller. (I did a series of three posts a while back about her 2008 article "Quirky Citizens: Autism, Gender and Reimagining Disability," which I thought made a lot of really good points). The more recent article, published in the summer 2009 issue of Signs, is called "The Geneticization of Autism: From New Reproductive Technologies to the Conception of Genetic Normalcy."

That odd word, "geneticization," gives you a clue as to the article's main premise: it implies treating autism as if it were genetic when it may or may not really be genetic.

Here is how Bumiller introduces the term and relates it to autism:

The term "geneticization" refers to the growth of genetics as a means to account for and explain health and disease and the process by which biological conditions constitute social definitions of normality and abnormality (Lippman 1991, 18). Abby Lippman coined this term in a feminist analysis of the growing influence of genetic determinism on public policies and private practices regarding pregnancy and health care and to emphasize the gender, race, and class implications of this trend. In particular, Lippman identified the need to study how genetic interventions affect health management in a variety of economic and social contexts. Feminist scholars have been wary of the coercive and normalizing power of medical professionals, yet they have also demonstrated the complex implications of biomedical advances. As Donna Haraway has persuasively argued, it makes little sense to be "simply oppositional" in response to this new technological future because we are deeply implicated in scientific progress (Haraway 1997, 3).

The shift in the autism field was first generated by the efforts of parents with autistic children, who were responding to regressive beliefs within the medical field. The scientific and popularized explanation for autism prior to the 1980s perpetuated a theory that pathological mothering was at the root of the disorder. Bruno Bettelheim (1979) is especially noted by critics for his view that childhood disturbances associated with autism did not arise spontaneously but resulted from extremely abnormal mother-child relations. Bettelheim's understanding of the condition is drawn from the seminal work of Leo Kanner (1943), who distinguished autism from schizophrenia as an innate or inborn disturbance of affective contact but also thought that the notable coldness and formality of the parents usually had some effect on the development of the condition. The hypothesis that autism can be attributed to a general lack of maternal warmth is called the "refrigerator mother" theory of autism, and it did not come under direct attack until Bernard Rimland criticized it in his 1964 book Infantile Autism: The Syndrome and Its Implication for a Neural Theory of Behavior. Both a parent of a child with autism and trained as a physician, Rimland undertook scientific work and activism that played a central role in recasting autism as a medical condition with distinct psychological symptoms that could potentially be remediated through diet and other therapies. In an era when parents of children with disabilities were beginning to organize and seek legitimacy, parents of autistic children embraced new biological explanations.

She goes on to describe a gradual narrowing of focus from "biological" explanations to specifically genetic ones, starting when researchers conducting twin studies found a strong pattern of heritability in autism. Later research identified a "broad autism phenotype" in parents and siblings of autistic children*, which added to the impression that it runs in families. But research into specific genes has failed to find much --- many candidate genes have been found, but each one only accounts for a tiny fraction of cases of autism. However, this review on the genetics of autism, published this month in Genetics in Medicine, estimates the total proportion of autistic people who have one of the genetic variations discovered so far at about 25%, which is actually a decent-sized chunk. Couple that with the relative newness of microarray-based comparative genomic hybridization --- the technique responsible for turning up a lot of these candidate genes --- and you see that it might be a bit premature to declare genetic research moribund**.

Kristin Bumiller thinks the continued assumption that autism must be a genetic condition has persisted beyond what the available evidence argues for, and that the idea that it may have environmental triggers is dismissed too readily. (I think the issue of whether the case for a genetic basis for autism is overstated or not is a bit more complicated; see above paragraph).

I do not argue with her assessment of the different political implications of genetic vs. environmental causes of developmental disability; if the former idea (i.e., that it's genetic) is widely believed, then each citizen is responsible for knowing hir own genetic status and making whatever reproductive choices follow from that status, while, if autism and other developmental disabilities are thought to be triggered by some environmental contaminant, then the responsibility falls to the government to enact stricter controls on neurotoxic pollutants. (Both of these scenarios presume a social context whose primary emphasis is on preventing, rather than accommodating, disability. In a less ableist society, deciding whether or not to have children if you're a carrier of [whatever] genes would be much less fraught with emotion and social pressures, although I think we'd want to limit pollution even if we weren't constantly told that developmental disability is a tragic waste of life).

Bumiller calls the individual-responsibility scenario "genetic citizenship": you become a member of a community of people affected by a given genetic disease, and that community advocates for research funding and participates in studies to help speed the development of cures or therapies:

The concept of genetic citizenship has been introduced to describe individuals in the age of biomedicalization who engage in a new style of activism related to their inheritable identities and differential embodiment (Heath, Rapp, and Taussig 2004). This concept is most frequently applied to situations in which individuals and family members affected by a genetic disease come together and take an active role in fundraising, advocating, and influencing scientists in the hope of finding a cure. In the past decade numerous disease-specific advocacy organizations have exercised significant influence over research priorities, affected capital allocation, sponsored gene banks, and demanded collaboration in the pursuit of real progress for people living with genetic diseases (Terry et al. 2007). This participation also takes advantage of new networking opportunities created by the Internet and the emergence of virtual communities. These citizens are seen as having cast off the role of passive patients to become active consumers of health services. As collectivities they have strived to maximize their influence on the development of new science, technology, and medical knowledge (Rose 2007, 23).

Autism advocacy provides an important vantage point from which to evaluate the presumed desirability of genetic citizenship because its activism is complicated by intense controversies about the significance of the genetic link and about the social identities of autistics. With the expansion of biomedical research in the field, much autism advocacy has shifted from promoting the well-being of affected families and children to searching for a cure. These new organizations, now consolidated under the banner of Autism Speaks, primarily promote biomedical research and are modeled on other fundraising campaigns that draw attention to the plight of people who suffer from rare diseases. ...
...
Prior studies of genetic citizenship have raised concerns about how the victories of new genetic movements may reflect a questionable convergence of individual interests and market forces and have noted how research priorities are driven by profit motives (Duster 2003a). Similar issues arise in the context of autism; much of this research is conducted in collaboration with high-profile genetic laboratories and large biotech corporations such as deCODE Genetics. This research is given priority despite uncertainty about whether and how genetic information will eventually be useful for pre- or postnatal genetic screening, diagnosis, or treatment methodologies. Both the irresolution about the potential benefits of genetic research and the oversimplifaction of its significance in the media diminish the power of consumers and the general public to either shift priorities or call for more transparency on the part of medical professionals. Moreover, the current focus on instrumental (and uncertain) goals rather than more broadly framed issues of social justice and welfare limits the role of disease-specific advocacy organizations in setting priorities (Stockdale 1999). Specifically in the case of autism awareness, it has been shown that since public discourse has focused on the medical paradigm --- particularly on efforts to find a cure --- there has been less focus on the rights and social welfare dimensions of the issue.

So there are good and bad things about this trend toward people organizing and cooperating with pharmaceutical and biotechnology companies to raise money and set research priorities. People are able to find other people --- whole communities of people --- who share their circumstances, and can thus pool their knowledge, resources and coping skills in ways they couldn't before, when each person essentially had to work everything out for hirself. That's a good thing; it's also a good thing that people affected by various conditions have a way to tell the scientific and medical communities what they need most.

What's not so good is the fact that the "partnership" between grassroots advocacy groups and corporations is so lopsided; this restricts disease-based activism to the corporate-philanthropic model of fundraising for biomedical research, when it might include that and campaigns for social changes geared toward creating a healthier society.

The other things Bumiller thinks are bad about the "genetic citizenship" model are its tendencies to coerce people into making certain health and reproductive choices, to exacerbate social and economic inequalities, and to put a heavy burden of responsibility on women as guardians of their families' health:

[D]isability activists have illuminated the implications of wide-scale genetic screening for the devaluing of disabled lives, particularly as the lines between state policy and individual choice are becoming increasingly blurred. Despite the well-established obligation of physicians and genetic counselors to provide nondirective advice, studies have found that patients were given information that imposes professionals' views about the usefulness of genetic knowledge and the parental responsibility to promote fetal health (Rapp 1999). This research has shown that professionals effectively delivered the message that the only rational choice is to give birth to a "normal" child. Genetic testing is now understood as a necessary component of preventative public health programs, largely as a result of the trend toward universalized testing and mandatory screening of newborns (van den Daele 2006). This shift has transformed prenatal testing from an option individual women are given to lower their risk of having a child with a genetic defect to a system of reducing overall health problems in the population (Ward 2002). ...
...
The research on prenatal counseling has also shown that professionals often frame genetic testing as necessary for socially responsible parenting. One study found that counselors presented prenatal testing as something women need in order to become good parents (Lippman 1991). In this sense, good parenting is about having the knowledge and resources provided by this testing and then following through in a socially responsible fashion. As genetic testing is fully incorporated as a standard of care for pregnant women, the act of refusal is no longer about the assumption of individual risk. Now, the noncompliant woman has failed to take advantage of an important opportunity to maximize the life chances of her child. Such actions are likely to be seen as contrary to good citizenship in an age of biopolitics, where the technologies of biomedicine have created a context in which "biology is not destiny, but opportunity" (Rose 2007, 51) and the desired course of action is to follow a strategy of life "optimization" (6). This strategy, according to Nikolas Rose, is "not eugenics but is shaped by forms of self-government imposed by the obligation of choice, the desire for self-fulfillment, and the wish of parents for the best lives for their children." He goes on to say that "its logics and its costs deserve analysis on their own terms" (69).

To see this as part of a new regime of choice is to fail to recognize the unintended consequences of life optimization in regard to the regulation of normalcy. A recent ethnographic study on the influence of new genetic knowledge on Belgian insurance companies aptly illustrates this dynamic at work. Ine Van Hoyweghen, Klasien Horstman, and Rita Schepers (2006) investigated how insurers take account of predictive medicine in the process of determining premiums. They describe the companies' decision making as a process of "making the normal deviant" because when insurers make judgments "the margin of being normal is actually quite small and the scope for deviation is quite wide" (Van Hoyweghen, Horstman, and Schepers 2006, 1229). They find that when insurers rate people with genetic predispositions they put extra emphasis on how they have managed their health. ... [I]nsurers impose a greater responsibility for optimally managing one's health on people with known risk factors. The authors' conclusions have serious implications for the social costs of life optimization: "Instead of a 'genetic determinism,' it seems more plausible that we are all subject to different levels of susceptibility. ... As a consequence, ... the individual's lifestyle habits, preventive initiatives and compliant behavior in relation to these susceptibilities could be stressed more" (Van Hoyweghen, Horstman, and Schepers 2006, 1233). The actuarial process imposes a norm that defines suitable lifestyles for people with risky genes. This creates an incentive system for genetic "deviants" to conform to normal expectations of proper lifestyles in order to satisfy social expectations.
...
The concurrent forces of life optimization under conditions of biomedicalization and demands for personal responsibility in a neoliberal welfare regime make the determination of a disabled person's worthiness central to the process of gaining public health resources. The rights afforded to people with disabilities are more available to those who are good genetic citizens and can demonstrate their strict compliance with social norms. For example, special education provisions rely on eligibility and service determinations that are individualized and ad hoc rather than derived straightforwardly from medical diagnosis. As a consequence, parents with poor genetic literacy often have trouble convincing schools that their children's behavior is the result of a biological condition rather than their bad choices as parents. In social security disability determinations, each case is processed according to subjective criteria used to measure a person's ability to work. Studies show that success in claiming disability depends on a person's ability and willingness to persevere through the application process (Bilder and Mechanic 2003). Since most claims are routinely denied and these denials lead to a lengthy appeals process, only those who are unusually skilled at conveying medical knowledge, or at enlisting the assistance of medical professionals, are likely to have their applications eventually approved.

These systematic processes have the effect of distinguishing between disabled people who are at low risk and those who are at high risk for becoming dependent on the state. Social policies that rely on dividing people up according to risk groups also cut against the organic sense of solidarity that develops among people with disabilities (or among their advocates and caretakers). These systems of classification rely on distinctions that are often contrary to a dynamic and inclusive sense of citizenship among people with disabilities. The overall effect of a person's genetic status interacting with other forms of inequality is to create conditions of "cumulative social and economic disadvantage" and consequently to reduce opportunities to participate in civic life (Kelly 2002, 181).

I don't think Bumiller is endorsing any kind of conspiracy theory, or that she believes some secret cabal of biotechnology executives ever held a meeting and drew up a plan to seize control of the emerging wave of health activism. Similarly, I don't think she thinks doctors and genetic counselors are trying to "regulat[e] normalcy"; both of those things just happened, as new technologies and social movements were assimilated into a corporate-dominated, individualistic society. No human masterminds or conspiracies --- indeed, any planning or intention at all --- need ever be involved.

These are all the things I agreed with in Bumiller's article; there's also some stuff I have problems with (besides my differing assessment of the state of research into the genetics of autism), which I will write about in a later post.

Bumiller, K. (2009). The Geneticization of Autism: From New Reproductive Technologies to the Conception of Genetic Normalcy Signs: Journal of Women in Culture and Society, 34 (4), 875-899 DOI: 10.1086/597130


*Not everyone who studied relatives of autistic children found this broad autism phenotype --- this 1994 study of 44 families with multiple autistic children found that the non-autistic siblings did not display any noticeable autistic-like traits; "in the vast majority of cases, children [were] either clearly affected or clearly unaffected". But from what I can see, most of the studies published on this topic did find evidence of a broad autism phenotype, although positive results may be overrepresented among published papers just because positive results are more often published than negative results.

**You could still criticize it on other grounds, from the potential for eugenic applications of genetic research to the questionable wisdom of using limited research funds to pursue such theoretical questions ("Where does autism come from?") rather than finding out what works to improve autistic people's lives and enable them to participate in society to a greater extent.

Testosterone and Estrogen Have Opposite Effects on Expression of a Gene Thought to be Underexpressed in Autism

Kev at Left Brain/Right Brain linked to this press release describing a new study published last week in PLoS ONE, that builds on several previous studies comparing gene expression in autistic people and their non-autistic siblings.

This study focuses on one gene, RORA, whose product is a steroid hormone receptor that also acts as a transcription factor. Previous studies had identified this gene as one that is expressed to a lesser degree in autistic people than in their non-autistic siblings; two studies compared the amount of mRNA (the product of DNA transcription, which serves as a template for protein synthesis) corresponding to certain genes that was present in cells taken from autistic people and their siblings, while a third study compared levels of DNA methylation (i.e., inactivation; methylated DNA is harder to transcribe) between those two groups. Only the third study mentions RORA, but the other two identify clusters of genes whose products do similar things, and are involved in similar processes, to RORA.

This study looks at the regulation of RORA by steroid hormones; its authors --- Drs. Valerie Hu and Ray-Chang Wu, along with grad students Terawit Sarachana and Minyi Xu --- found that both estrogen and testosterone affect its functioning (as assessed by bathing cultures of cells with solutions of varying strengths of either hormone, and then using qRT-PCR to determine how much RORA mRNA was produced under each condition), and that their effects are directly opposed to one another: testosterone suppresses the transcription of RORA, while estrogen enhances it.

They also found sequences within another gene --- that for aromatase, an enzyme that converts testosterone into estrogen --- that are the same as those making up RORA binding sites in other genes, which indicates that RORA might be involved in regulating the production of aromatase.

Given this, Sarachana and colleagues see the potential for a feedback loop: in the presence of testosterone, RORA is less active, which could mean that less aromatase is produced, which would mean more testosterone accumulates, leading to even more suppression of RORA.

Figure 5, in Sarachana et al., 2011 --- diagram showing the up- or down-regulation of RORA by the steroid hormones estrogen and testosterone, and the regulation of aromatase, and therefore estrogen synthesis, by RORA

(There's also a potential feedback loop involving estrogen, which up-regulates RORA, which up-regulates aromatase, which in turn produces more estrogen).

There was an upper limit to the dose-response effect of either hormone, though: for both, it was the 1-nanomole/L solution that had the strongest effect on RORA transcription, which seems to suggest that at very high concentrations of testosterone, RORA transcription might creep back upward.
Figure 1, in Sarachana et al., 2011; graphs showing the relative amounts of RORA expressed when cells are treated with three different strengths of a solution of either dihydrotestosterone (C) or 17-beta-estradiol (A), and also changes in expression of RORA over time following treatment with 1-nM solution of DHT (D) or estradiol (B)

Given that RORA has already been found to be less active in autistic people than it is in their non-autistic siblings, the suppression of RORA by testosterone provides a possible biological mechanism underpinning the greater prevalence of autism among males than females, and also associations found between higher levels of testosterone exposure in utero and later development of autistic traits.

There is some evidence for this proposed relationship; Sarachana and colleagues also assayed(post-mortem) brain tissue from autistic (n = 12) and non-autistic donors (n = 22), measuring both RORA and aromatase levels by immunofluorescence. They found two things that would tend to support their low RORA --> low aromatase --> more testosterone hypothesis: RORA is indeed present in smaller amounts in the brain tissue from autistic donors, and aromatase levels are strongly correlated with RORA levels in all the brains that were analyzed.
Figure 4A, in Sarachana et al., 2011; comparison between representative samples of brain tissue labeled to show either RORA (red) or aromatase (blue) expression. The top two samples come from female donors; the bottom two come from male donors. The second and fourth sample (one male, one female) come from autistic donors; the first and third come from age-matched non-autistic donors. You can see that the samples from the autistic donors are dimmer than the control samples; less color indicates less of the compound being assayed for, since the fluorescent molecule is attached to an antibody that will only bind to the protein of interest (RORA or aromatase, in this case).
Figure 4C, in Sarachana et al, 2011; graph showing the relationship, R², between the concentration of RORA and the concentration of aromatase present in tissue samples from both autistic and non-autistic donors

Now, if you're like me, you'll be wondering how strong the evidence is that a non-trivial proportion of autistic people actually do underexpress RORA. While the immunofluorescence assay of the tissue samples didn't falsify the hypothesis that autistic people's brain cells would produce less RORA and aromatase, because it involved only 34 brains, its findings can't be extrapolated to autistic people in general (or even to a specific subgroup of autistic people in general) without gene-expression data from a larger, more representative autistic sample.

In this article, Valerie Hu and her co-author Mara Steinberg describe a new method for sorting autistic study participants into subgroups based on how they scored on individual questions on the Autism Diagnostic Interview (ADI-R). The rationale for this, which Jon Brock has elegantly demonstrated on his blog, is that two people can both meet diagnostic criteria for autism without having even a single trait in common. So Hu and Steinberg's method is supposed to sort autistic people into groups with the people whose clinical profiles most match their own.

They came up with four categories: people with severe language impairment, people with savant skills, people with relatively mild symptoms distributed across all the ADI-R domains (i.e., language, nonverbal communication, social interaction, play skills, interests and behaviors, physical sensitivities and mannerisms, aggression, and savant skills), and people with intermediate (i.e., neither mild nor severe) symptoms who were also less likely than the other groups to have savant skills. Three of these groups were chosen for inclusion in later gene-expression studies: the language-impaired group, the savant group, and the mild-autism group. Other people weeded out of the gene-expression study population were people whose autism was part of a genetic syndrome, like Rett syndrome or Fragile X, people with other psychiatric diagnoses, people who were born prematurely, people with significant cognitive impairment, and female people (!).

(I suspect the last group --- women and girls with autism --- were left out just because there were so few of them. Of the 1,954 people who were sorted into groups, only 428 were female.)

Once all the potential confounders were weeded out, the researchers had cell lines derived from 86 autistic individuals --- 31 in the language-impairment group, 26 in the mild-autism group, 16 in the savant group, and 13 who apparently met inclusion criteria for both the savant and language-impairment groups --- which made up the autistic sample in this study, contrasted with 30 non-autistic controls. The other study Sarachana and colleagues cite as providing evidence of RORA underexpression in autistic people is this one, which predates Hu and Steinberg's method for dividing autistic people into subgroups; its population, since it is composed of identical twins, is a lot smaller: just six sets of twins (five with at least one twin diagnosed with autism) and non-autistic siblings of two of the sets of twins --- fourteen people overall.

So, while the model Sarachana and colleagues propose --- a predisposition for autistic people to suppress expression of RORA, which both creates and is reinforced by a high-androgen environment --- is really interesting, and could shed light on 1) how male and female autistics differ and 2) what the male:female ratio within autism might be in a perfect world where diagnostic criteria weren't geared more toward men and boys, I'd like to see the underexpression-of-RORA finding replicated in a bigger study.

Sources:
Hu, V., Frank, B., Heine, S., Lee, N., & Quackenbush, J. (2006). Gene expression profiling of lymphoblastoid cell lines from monozygotic twins discordant in severity of autism reveals differential regulation of neurologically relevant genes BMC Genomics, 7 (1) DOI: 10.1186/1471-2164-7-118

Hu, V., Sarachana, T., Kim, K., Nguyen, A., Kulkarni, S., Steinberg, M., Luu, T., Lai, Y., & Lee, N. (2009). Gene expression profiling differentiates autism case-controls and phenotypic variants of autism spectrum disorders: evidence for circadian rhythm dysfunction in severe autism Autism Research, 2 (2), 78-97 DOI: 10.1002/aur.73

Hu, V., & Steinberg, M. (2009). Novel clustering of items from the Autism Diagnostic Interview-Revised to define phenotypes within autism spectrum disorders Autism Research, 2 (2), 67-77 DOI: 10.1002/aur.72

Sarachana, T., Xu, M., Wu, R., & Hu, V. (2011). Sex Hormones in Autism: Androgens and Estrogens Differentially and Reciprocally Regulate RORA, a Novel Candidate Gene for Autism PLoS ONE, 6 (2) DOI: 10.1371/journal.pone.0017116