nature biotechnology • VOLUME 19 • MAY 2001 • http://biotech.nature.com

NEWS AND VIEWS

418

the growth of strategy II plants in soils withlow iron availability, what about strategy Iplants? It is possible that increased synthe-sis of nicotianamine might confer somebenefit, but more likely gains may comefrom increasing the rate-limiting step ofFe(III) chelate reduction. And interestingly,strategy I plants may also benefit from therelease of chelators. Plants engineered torelease more citrate into the rhizospheregrow better than nonengineered plants insoils of low iron availability (Luis Herrera-Estrella, personal communication).Presumably, citrate forms complexes withFe(III) that can then be reduced at the rootsurface for transport across the plasmamembrane.

In addition to improvements in theuptake of iron from the soil, attention isbeing paid to increasing the storage formsof iron. Rice has been engineered to havehigher levels of the iron storage protein fer-ritin in the grain, but results conflict as tothe effect of such overexpression on theiron content of the grain11,12. And none ofthe rice plants, including those described byTakahashi et al., has been tested in the field.

The idea of fortifying foods with iron isnot new. Iron-fortified infant formula andiron-enriched flours are widely available.But by fortifying food crops before harvest,we stand the best chance of ensuring thatsuch iron-fortified foods get to the peoplewho most need them. The engineering ofrice endosperm to produce more vitamin Aby Ingo Potrykus, Peter Beyer, and their co-workers13 is the gold standard for preharvestfortification of rice to increase its nutritionalvalue. The goal of producing rice with bothmore vitamin A and more iron in its grain iscertainly within reach. Such rice would pro-vide a solution for two of the three majornutrient deficiency problems facing theworld today.

Sixty to eighty percent of the world’s pop-ulation may be iron deficient14. Iron-deficiency anemia is responsible for an esti-mated 20% of maternal deaths in Africa andAsia, and can lead to irreversiblephysical/mental retardation, reduced resis-tance to infection, and reduced work perfor-mance. The approach described byTakahashi et al. has a dramatic effect onplant yield. We now eagerly await confirma-tion that it also bolsters iron content.

1. Marschner, H. Mineral nutrition of higher plants.(Academic Press, Boston; 1995).

2. Takahashi, M. et al. Nat. Biotechnol. 19, 466–469(2001).

3. Robinson, N.J. et al. Nature 397, 694–697 (1999).4. Eide, D. et al. Proc. Natl. Acad. Sci. USA 93,

5624–5628 (1996).5. Curie, C. et al. Nature 409, 346–349 (2001).6. Takahashi, M. et al. Plant Physiol. 121, 947–956

(1999).

7. Higuchi, K. et al. Plant Physiol. 119, 471–479(1999).

8. Herbik, A. et al. Eur. J. Biochem. 265, 231–239(1999).

9. Ling, H. Q. et al. Proc. Natl. Acad. Sci. USA 96,7098–7103 (1999).

10. von Wiren, N. et al. Plant Physiol. 119, 1107–1114

(1999).11. Goto, F. et al. Nat. Biotechnol. 17, 282–286

(1999).12. Drakakaki, G. et al. Transgenic Res. 9, 445–452

(2000).13. Ye, X. et al. Science 287, 303–305 (2000).14. http://www.who.int/nut/ida. WHO (2001).

Twenty-three research groupsfunded by microarray initiatives atthe National Institute on DrugAbuse (NIDA)1 and the NationalInstitute of Diabetes and Digestiveand Kidney Diseases (NIDDK)2 ofthe US National Institutes ofHealth (NIH; Rockville, MD)recently took part in a meeting* todiscuss current logistical chal-lenges in microarray research andformulate a strategy for facilitat-ing progress in the field. Afterextensive discussion, a plan wasproposed incorporating the estab-lishment of effective communica-tion tools, use of a standard subsetof several hundred genes, adop-tion of a single gene-naming pro-tocol, design of a set of standardquality controls, and the development of acontrol pool of RNA, together with the req-uisite bioinformatics tools. The USNational Center for BiotechnologyInformation (NCBI; Rockville, MD) GeneExpression Omnibus (GEO) was suggestedas an ideal repository for microarray data.We believe that such an arrangement couldserve as a model for other national/interna-tional microarray data curation/storageefforts.

Biologists attempting microarray analysisof gene expression currently face severalchallenges, not least the rapid evolution ofarray technology, methodology, statisticaland bioinformatics analysis tools, and theabsence of widely accepted quality controls.

Although the belief that “more data are bet-ter” is common, biologists are experiencingdifficulty in keeping up with the latestadvances, analyzing data, and comparingdata obtained using competing technologiesor at other laboratories. In addition, thefield lacks uniform standards to describesimple (e.g., gene names) and complex (e.g.,experimental details, analysis methodology)parameters.

The Rockville meeting was convened todiscuss these issues and find a consensus forcoordinating research. During the first day,participants discussed recent advances inexperimental design, methodology, and dataanalysis. Several speakers described method-ological and analytical approaches forreducing and measuring experimental error.Peter Munson from NIH’s Center forInformation Technology discussed theimportance of replicates in experimentaldesign and the use of simple statistical toolsto both assess the quality of data and distin-guish outliers from the intrinsic error of theexperiment. Michael Miles from the GalloCenter at the University of California in SanFrancisco outlined normalization approach-es, including statistical methods such as pin-

Searching for array standards in RockvilleA recent US NIH meeting produced a draft pilot protocol to facilitatemicroarray data sharing.

Robert A. Star and Rebekah S. Rasooly

Robert A. Star is senior scientific advisor andchief, renal diagnostics and therapeutics,Division of Kidney, Urologic, and HematologicDiseases, NIDDK/NIH, and Rebekah S.Rasooly is program director, cell biology &genetics at the Division of Neuroscience &Behavioral Research, National Institute onDrug Abuse/NIH, Bethesda, MD 20892-9555(rrasooly@nida.nih.gov).

Figure 1. Brainstorming in the NIH Neuroscience Center inRockville, MD. Twenty three US scientists from the NIDAand NIDDKA met at the NIH earlier this year to hammer outprotocols for facilitating array data sharing and analysis.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://b

iote

ch.n

atu

re.c

om

© 2001 Nature Publishing Group http://biotech.nature.com

NEWS AND VIEWS

http://biotech.nature.com • MAY 2001 • VOLUME 19 • nature biotechnology 419

specific normalization, which can assess theoverall quality of the array experimentbefore looking for particular patterns ofexpression.

Error and background can also bereduced experimentally, according to SteveGullans from Brigham & Women’s Hospitalin Boston, through the use of duplicatespots, thorough mixing, and red–greenlabel reversal. In his talk, Gullans also dis-cussed statistical considerations in normal-izing data and described a “shared gene set”of a few hundred genes that he is using fornormalization and will make available toothers soon. An approach termed den-drimer signal amplification that reduces theamount of RNA needed for each array3 wasmentioned.

Two talks emphasized the challenges ofcreating repositories for microarray data.Scott Markel, from NetGenics (Cleveland,OH) and co-chair of the Life SciencesResearch Domain Taskforce of the ObjectManagement Group (OMG), describedOMG’s efforts to foster the development ofa single software standard for presentingmicroarray data. OMG is working towardadopting a standard by the summer of 2001,and he invited the microarray research com-munity to participate in this effort. AlexLash from NIH’s NCBI introduced GEO, anew public repository for data from large-scale expression assays such as microarrays.GEO’s approach to the existing diversity ofdata formats is to impose very few require-ments for submissions to GEO. Lash report-ed that advanced query features are beingdeveloped to allow customized searches ofthe repository.

During the second day, separate breakoutmeetings of NIDA and NIDDK granteesrevealed several common points. First, webpages and listserves were highlighted as use-ful fora for communication and “shoptalk”concerning microarray data. Toward thisend, the NIDDK group is building a websiteat http://www.niddk.org/ that will containboth “best practice” and “dud” protocols forindividual array platforms and tissues andlinks to other microarray sites. Second, a“shared gene set” of human genes, asdescribed by Steve Gullans, should be adopt-ed whenever practical. The expression ofgenes chosen for such a set would vary con-siderably from tissue to tissue, but all wouldbe expressed in all tissues, thereby facilitat-ing quality control and data sharing. Third, acommon strategy should be adopted fornaming genes, relating oligonucleotide genechip data to complementary DNA (cDNA)clones, and target gene to existing data inGenBank and other repositories. Finally, apool of control RNA should be developed toenable data standardization and quality

assessment. Comparing the control poolwith an individual laboratory’s control sam-ple should facilitate data sharing betweenlaboratories and help each research groupmonitor and compare experiments overtime. Toward this end, participants at themeeting from NIDA agreed to develop apool of rodent brain RNA that will be dis-tributed by a single, central laboratory.NIDDK-funded laboratories agreed to testsamples from human and mouse RNA poolsand perform a “same–sample control” byconducting the same experiment withreversed labeling.

NIDA grantees also agreed to store com-plete published and unpublished data sets inGEO within one year after the end of eachgrant project period. A NIDA working groupwill develop the format for GEO submis-sions, specifying how genes will be namedand which quality control measurementsand normalization details will be included.In addition, the group agreed to develop alist of the experimental details and descrip-tions, to be provided for each experiment(e.g., drug used, dose, animal species, strain,and age). Finally, investigators with core setsof genes relevant to addiction/neurobiologywere encouraged to make those cDNAclones available to others.

The NIDDK grantees also agreed to worktoward improving the availability and quali-ty of DNA intended for chips, and validatinga list of commercially available sequence-verified cDNAs that can be used by all labo-ratories. A working group will be set up toexplore the potential to develop longoligonucleotides specifically for genesexpressed in the pancreas, kidney, bladder,and other organs of interest.

Overall, the meeting achieved two goals.First, participants gained a sense of the tech-nical challenges involved in producing high-quality microarray data and drafted thebeginnings of protocol and data qualitystandards to facilitate data sharing amonglaboratories. Second, NIDA and NIDDKsponsors now have a framework for assem-bling microarray data, which will allow thebroader research community to analyzemultiple data sets that explore the same basicresearch question. These goals will be essen-tial for both increasing the quality of dataand facilitating the exchange of microarraydata among and between laboratories.

*First Annual Meeting of NIDA and NIDDK RFA-Funded Microarray Investigators, January11–12, 2001, Rockville, MD.

1. http://grants.nih.gov/grants/guide/rfa-files/RFA-DA-00-003.html.

2. http://grants.nih.gov/grants/guide/rfa-files/RFA-DK-00-002.html.

3. Stears, R. L., Getts, R.C., & Gullans, S.R. Physiol.Genomics 9, 93–99 (2000).

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://b

iote

ch.n

atu

re.c

om

© 2001 Nature Publishing Group http://biotech.nature.com