The manuscript by Glatt et al. raises important questions: Can combining data from blood and brain microarray studies uncover biomarkers of schizophrenia? Are the putative biomarker candidates identified in a combined blood-brain data set any better than the ones identified in peripheral biomarker studies? According to the authors, their approach will “…facilitate the discovery of highly reliable and reproducible candidate risk genes and biomarkers for SZ.” Unfortunately, previous attempts in identifying such peripheral markers were mostly unsuccessful, yet high-throughput expression profiling methods may have a chance to identify the proverbial needle in the haystack. However, only future replications of these findings will resolve if the approach presented in this manuscript represents a breakthrough in the search for peripheral schizophrenia biomarkers.

The authors employ an intriguing and novel data-mining strategy using CORGON. Unfortunately, while this software has produced improved false discovery in non-neural data sets, we have little information on its...  Read more

The manuscript by Glatt et al. raises important questions: Can combining data from blood and brain microarray studies uncover biomarkers of schizophrenia? Are the putative biomarker candidates identified in a combined blood-brain data set any better than the ones identified in peripheral biomarker studies? According to the authors, their approach will “…facilitate the discovery of highly reliable and reproducible candidate risk genes and biomarkers for SZ.” Unfortunately, previous attempts in identifying such peripheral markers were mostly unsuccessful, yet high-throughput expression profiling methods may have a chance to identify the proverbial needle in the haystack. However, only future replications of these findings will resolve if the approach presented in this manuscript represents a breakthrough in the search for peripheral schizophrenia biomarkers.

The authors employ an intriguing and novel data-mining strategy using CORGON. Unfortunately, while this software has produced improved false discovery in non-neural data sets, we have little information on its performance in brain microarray data sets, especially in comparison to the more mainstream gcRMA, PLIER, or SAM. Similarly, we have also little information on the Gene Ontology analysis software used (MADCAP, referenced only by a conference proceeding from 2003), and how does it compare to the standard GO mining algorithms (e.g., DAVID/EASE, developed by NIAID)? There has been an impressive body of postmortem schizophrenia microarray literature generated over the last 5 years. These microarray data sets gave rise to many common observations across cohorts and platforms, but none of this work is discussed or referenced in this manuscript. Failing to confirm at least some of the previous findings of glial, synaptic, or metabolic expression changes in the schizophrenic brain tissue is a concern, and it can be a consequence of experimental design, cohort differences, or the analytical approaches employed. However, if the strategy of the present study did not confirm previously replicated expression differences in the brain, it is unclear how it can be used to search for overlapping markers between the brain and the blood.

The study by Ming Tsuang and colleagues reporting increased SELENBP1 in schizophrenia is very interesting. Might that result in reduced function of the recently described anti-inflammatory gene selenoprotein S on chromosome 15 (1)? Curran et al. report that suppression of SEPS1 results in increased release of TNFα and Il-6. Gilmore et al. (2) report that cytokines generated in response to infection, IL-1β, TNFα, and IL-6 can significantly reduce dendrite development and complexity of developing cortical neurons, consistent with the neuropathology of schizophrenia.

References:
1. Curran JE, Jowett JB, Elliott KS, Gao Y, Gluschenko K, Wang J, Azim DM, Cai G, Mahaney MC, Comuzzie AG, Dyer TD, Walder KR, Zimmet P, Maccluer JW, Collier GR, Kissebah AH, Blangero J. Genetic variation in selenoprotein S influences inflammatory response. Nat Genet. 2005 Oct 9; [Epub ahead of print] Abstract

2. Gilmore JH, Fredrik Jarskog L, Vadlamudi S, Lauder JM. Prenatal Infection and Risk for Schizophrenia: IL-1beta, IL-6, and TNFalpha Inhibit Cortical Neuron Dendrite Development. Neuropsychopharmacology. 2004 Jul;29(7):1221-9. Abstract

View all comments by Mary Reid

I wonder whether the study by Ishida et al. (1) finding that acetaminophen cytotoxicity is enhanced in selenium-binding protein-overexpressed COS-1 cells may help explain the results reported by the Mortensen group in two separate studies. In one, they find that mothers who take aspirin and acetaminophen during the second trimester have a fourfold risk of giving birth to a child who later develops schizophrenia as an adult (2). In the other, they report an increased risk of schizophrenia following paracetamol poisoning (3). It would seem that should SELENBP1 be a valid marker for the diagnosis of schizophrenia, then paracetamol would be contraindicated.

A further study by Ishida and colleagues (4) reports induction of hepatic selenium-binding protein by aryl hydrocarbon receptor ligands in rats. Does this suggest that AHR ligands such as polycyclic aromatic hydrocarbons and polychlorinated dioxin compounds may cause schizophrenia? Does this explain the reports of schizophrenia in the Vietnam veterans?

References:
1. Ishida T, Abe M, Oguri K, Yamada H. Enhancement of acetaminophen cytotoxicity in selenium-binding protein-overexpressed COS-1 cells. Drug Metab Pharmacokinet. 2004 Aug;19(4):290-6. Abstract

2. Sorensen HJ, Mortensen EL, Reinisch JM, Mednick SA. Association between prenatal exposure to analgesics and risk of schizophrenia. Br J Psychiatry. 2004 Nov;185:366-71. Abstract

3. Jepsen P, Qin P, Norgard B, Agerbo E, Mortensen PB, Vilstrup H, Sorensen HT. The association between admission for poisoning with paracetamol or other weak analgesics and subsequent admission for psychiatric disorder: a Danish nationwide case-control study. Aliment Pharmacol Ther. 2005 Oct 1;22(7):645-51. Abstract

4. Ishida, T., Ishii, Y., Yamada, H., and Oguri, K. The induction of hepatic selenium-binding protein by aryl hydrocarbon (Ah)-receptor ligands in rats. J. Health Sci., 48: 62-68, 2002.

View all comments by Mary Reid

Response to comment by Mary Reid posted 21 October 2005
I wish we knew more of the function of SELENBP1, other than the fact that it binds selenium. Does it sequester it, thus making it unavailable for biological activity, or is it a transporter or receptor for selenium, making it more active/available? Unfortunately, the ontology of the gene is poorly documented. One distinct possibility is that the role of the gene/protein is better understood, but under a pseudonym, such as heat shock protein 56 (HSP56). We need to do more reading into this possibility to determine what the role of the protein is, but we do know that it is expressed in the growing tips of neurites.

View all comments by Stephen J. Glatt

Response to comment by Karoly Mirnics posted 21 October 2005
Replication will be key, and we hope that others will utilize gene expression microarrays to search for potential schizophrenia biomarkers as we have done. Our findings of dysregulation of specific myelin-related genes are consistent with prior work, as is the energy pathway dysfunction. Our work in conjunction with the others is painting a consistent picture of dysfunction in schizophrenia.

View all comments by Stephen J. Glatt

It certainly will be of interest to see further studies regarding the function of SELENBP1. Porat et al. find that 56SBP participates in late-stage intra-Golgi transport, but suggest that it may have more than one physiological role. The authors propose that SBP56 is active downstream of the Rab proteins. Might it be worthwhile to look at Rab36-22q11.2? A study by Mori et al. localized Rab36 at the Golgi body, suggesting that it is involved in vesicular transport around the Golgi apparatus.

References:
Porat A, Sagiv Y, Elazar Z. A 56-kDa selenium-binding protein participates in intra-Golgi protein transport. J Biol Chem. 2000 May 12;275(19):14457-65. Abstract

Mori T, Fukuda Y, Kuroda H, Matsumura T, Ota S, Sugimoto T, Nakamura Y, Inazawa J. Cloning and characterization of a novel Rab-family gene, Rab36, within the region at 22q11.2 that is homozygously deleted in malignant rhabdoid tumors. Biochem Biophys Res Commun. 1999 Jan 27;254(3):594-600. Abstract

View all comments by Mary Reid

The paper by Glatt and coworkers reports on the gene expression differences in postmortem brain samples from individuals with schizophrenia and nonpsychiatric control samples. The researchers then compared these profiles with expression differences observed in blood from a separate sample of schizophrenic and nonpsychiatric controls.

Since there is no gold standard for the statistical evaluation of microarray data analysis, many alternative methods have been reported. In this study, Glatt et al. adopted their original method termed CORGON. One of the features of their method is treatment of noise. They assumed multiplicative rather than additive noise. Though it is difficult to evaluate the possible advantageous effect of assuming multiplicative rather than additive noise on the data set presented, analytical models which assume that the error of the observed expression signals of microarray is multiplicative have been well studied recently. Multiplicative models appear to be more suitable than simple additive models, at least for some microarray data sets. The...  Read more

The paper by Glatt and coworkers reports on the gene expression differences in postmortem brain samples from individuals with schizophrenia and nonpsychiatric control samples. The researchers then compared these profiles with expression differences observed in blood from a separate sample of schizophrenic and nonpsychiatric controls.

Since there is no gold standard for the statistical evaluation of microarray data analysis, many alternative methods have been reported. In this study, Glatt et al. adopted their original method termed CORGON. One of the features of their method is treatment of noise. They assumed multiplicative rather than additive noise. Though it is difficult to evaluate the possible advantageous effect of assuming multiplicative rather than additive noise on the data set presented, analytical models which assume that the error of the observed expression signals of microarray is multiplicative have been well studied recently. Multiplicative models appear to be more suitable than simple additive models, at least for some microarray data sets. The permutation test used to identify differentially expressed genes is not dependent on typical distributions used in general statistical analyses, but based on the distribution generated from observed values. It may be contradictory that they used the t-statistic in order to compare each permutation with the absolute value for unpermuted t-statistic, because it is well known that the statistic assuming univariate normal distribution, such as the t- statistic, shows poor sensitivity in detecting qualitatively consistent changes when it shows large quantitative variability. It is not clear if this problem can be avoided by using log expression values, which were adopted in this study.

It is exciting that SELENBP1 was discovered as a candidate biomarker for schizophrenia in common with both brain and blood. One of the issues to be validated is the reproducibility of the data by independent samples. The authors used blood samples to compare expression changes with brain samples, and found that six genes showed differential expression between cases and controls in these two apparently unrelated tissues. Perhaps one needs to be cautious, however, in the interpretation of these findings, since only one, SELENBP1, showed change in the same direction. It is not obvious how such expression changes in opposite directions reflect a common biological mechanism. The authors also demonstrated differential SELENBP1 protein expression in brain; however, being such a small (1.16-fold) mRNA expression difference, it is difficult to draw any conclusions about a possible correlation. Further, as shown in Figure 1 of the paper, the SELENBP1 staining was increased in glia, whereas the protein was decreased in neurons in schizophrenic brain compared with controls.

Response to comment by Graff and Kimura
We agree that it is not obvious how such expression changes in opposite directions reflect a common biological mechanism. This is precisely why we focused our subsequent efforts on SELENBP1. It is also important to realize that perhaps many more genes had corresponding differential expression in the blood and brain, but we may not have detected them due to the highly conservative nature of our statistical analyses.

We would also underscore that we don't know what fold-change difference between groups is biologically meaningful, which is why we focus on the significance level instead. The change in SELENBP1 may not be large, but it seems to be highly reliable. Replication efforts should prove very helpful in establishing the utility of SELENBP1 to identify schizophrenia.

View all comments by Stephen J. Glatt