The microarray and gene chip grew out of efforts by a team of scientists concerned with optimizing methods of drug discovery. This group was assembled by Alex Zaffaroni, the legendary CEO of Syntex and later founder of several biotech firms, including Alza and DNAX. In 1988 Zaffaroni approached Lubert Stryer, professor of biochemistry at Stanford and inventor of numerous fluorescence-tagging methods for enabling the Fluorescence Activated Cell Sorter (FACS) as one of the primary tools of cell biological research, to become the chief scientific officer of the new company Zaffaroni, J. Leighton Read, and Peter Schultz [Note A] were founding called Affymax. The goal of Affymax was to develop novel chemical approaches to automated drug discovery. The traditional approach in drug discovery had been to synthesize or discover new candidate drugs and then test their activities one at a time. This is a tedious, cumbersome, and increasingly expensive approach, so speeding up or automating this process was of substantial interest to pharmaceutical companies.

In building the company, Zaffaroni, Schultz, and Read did not have a specific technology they intended to pursue. However, feeling that recent developments in biotechnology were about to render the problem of drug discovery tractable, they assembled a star-studded scientific advisory board from Stanford and several other universities [Note B]. From the beginning the approach advocated by Avram Goldstein of the Stanford Pharmacology Department seemed most appealing. Goldstein urged the pursuit of peptide synthesis as a means of generating chemical diversity for identifying promising leads for drug molecules. Goldstein argued that since receptors for any ligand can be formed from short peptide sequences in the combining sites of antibodies, it must be conversely true that from short peptides, one could make a ligand for any receptor 6. Affymax could pursue the generation of large libraries of small peptides with novel sequences against various protein targets, analogous to the way in which the immune system operates by mass screening its antibody repertoire, identifying the ones that work best and making more of those.

Several methods for generating large peptide libraries through what was being called "combinatorial chemistry" were coming on the scene in the mid-1980s. The field actually got its start in 1963, when R. Bruce Merrifield (Nobel Prize in chemistry, 1984) introduced the concept of solid phase peptide synthesis (SPPS) whereby polypeptide chains as short as two amino acids (dipeptides) as well as longer (protein) chains could be made in assembly-line fashion using automated peptide synthesizers. In the 1980s, Australian researcher Mario Geysen of the University of Melbourne (later at Glaxo Wellcome) showed that SPPS could be the basis of multiple peptide synthesis. Geysen's "peptide on a pin" method generated a variety of short protein fragments by combining multiple amino acids (the building blocks of peptides and proteins) in different permutations 78. Each peptide was made on the end of a pin-shaped polyethylene support, dipped into a dish with a new amino acid for each step in the reaction. By lining the pins in an array with the (originally 96) wells of a microtitre plate dozens (even hundreds) of reactions could be performed at the same time. Geysen's method was the first example of a library of synthesized compounds where the molecular identity could be known based on the physical position of the compound in the library [Note C].

Other candidate techniques for generating combinatorial libraries of peptides were coming on the scene at about the same time the Affymax board was developing its approach [Note D]. But rather than pursuing any of these options in developing combinatorial syntheses, Read and Pirrung came up with a brilliant new approach of their own which they called VLSIPS, for

V

L

S

I

P

S

The next step for the group was to begin work on implementing the idea of generating chemical diversity on an array designed by a photolithographic process. Pirrung was about to head off to Duke University to take up a new professorship in biochemistry, so Stryer began inquiring among local colleagues for the name of a young biochemist who might be appropriate to head up the project of producing a prototype and reducing the invention to practice. Stryer's long-time Berkeley collaborator Alexander Glazer suggested Stephen Fodor, a young Princeton Ph.D. with a NIH postdoctoral fellowship working on time-resolved spectroscopy of bacterial and plant pigments in his lab. Glazer recommended Fodor as a biochemist of exceptional ability; indeed, he already had the reputation of a visionary. Although taking a position in industry was not of interest to Fodor, the opportunity to brainstorm with Zaffaroni, Stryer, Berg, Schultz, Lederberg and Davis was an opportunity he did not want to miss. The academic appointment could follow. In July of 1989 Fodor joined the group at the Affymax offices on Porter Drive in the Stanford Industrial Park.

Over the next 18 months Fodor worked intensely with Stryer in what both describe as the most stimulating and productive period of scientific invention imaginable. The invention they, together with their scientific colleagues at Affymax, ultimately produced – light-directed spatially addressable chemical synthesis – was quite literally the marriage of biochemistry and the photolithography techniques used in chip design in the local semiconductor industry. What they demonstrated by 1991 in a now classic article published in Science was a process for depositing onto a glass substrate – literally a microscope slide cover in the first version of the invention – amino acid groups – NH – that were blocked by a photolabile protecting chemical group – X (Figure 1) 9. Illumination with a laser through a mask led to photodeprotection, allowing, in the next step through chemical coupling, the addition of a first chemical building block A containing a photolabile protecting group X. In the next step, a different mask is used to photoactivate a different region of the substrate. A second labeled group B is then attached to the amino groups exposed by the illumination through the mask. This process is repeated as many times as desired to obtain the desired set of products. By washing a bath of peptide chains with a fluorescent marker attached to the end, it was possible to determine the composition of amino acids forming the chain with the aid of a photomultiplier/scanner that operated similarly to the FACS (Fluorescence Activated Cell Sorter). The initial microarray consisted of 1024 peptides in a 1.6 cm² area generated in a ten-step process. This was the first microarray designed specifically for peptide synthesis, and at the same time Fodor developed a scanner for reading the output.

Part of the beauty of combining photolithography with combinatorial chemistry is the resultant high density of the compounds on the substrate. Theoretically, the only physical limitation on the density is the degree to which the compounds can be activated – in other words, the diffraction of light. This provides for an incredibly high degree of miniaturization, and in 1991, at the time of the publication of the paper, Fodor and his colleagues at Affymax wrote:

Our present capability for high-contrast photodeprotection is better than 20 μm, which gives >250,000 synthesis sites per square centimeter. There is no physical reason why higher densities of synthesis sites cannot be achieved 10.

Using photolithography essentially brought Moore's Law to Affymax and to the new company Fodor launched around it, Affymetrix. Similarly to the semiconductor industry, Affymetrix has steadily increased the density of synthesis sites, while making the chips more complex and harder to manufacture 11. Indeed, five years later, Affymetrix had produced a prototype chip with a million probes 12.

While work continued on peptide microarrays, Stryer and the Affymax scientific board recognized a much more immediate opportunity in the development of nucleic acid microarrays. Solid phase synthesis of DNA is considered the most effective and reliable method of chemical synthesis known. Given the strict base-pairing rules (Watson-Crick pairing) obeyed by the four building blocks of DNA (adenine, cytosine, guanine, and thymine, or A, C, G, and T), a section of single-stranded DNA, which might contain numerous genes and thus be used as a probe, will match up only with its complementary strand of DNA [cDNA for "complementary DNA"] to form the double helix. RNA, which is DNA's chemical cousin, also follows a strict base-pairing rule when binding to DNA, so the sequence of any RNA strand that pairs up with DNA on a microarray can be inferred as well.

Combinatorial analysis based on light-directed synthesis of DNA on a chip offered excellent opportunities, and there were a number of reasons why Fodor wanted to pursue DNA chips more vigorously than peptide arrays. Simply in terms of practical considerations of construction, a peptide array of just two amino acid units for each sequence with the 20 amino acids as building blocks produces an array of 400 sequences in 40 steps of the sort described above. By contrast, in the same number of steps (40), a DNA array of 10 unit (ATTGC...) sequences each synthesized from the four nucleic acids as building blocks can be constructed containing an array of one million sequences. Moreover, DNA was ideally suited for light-directed synthesis, and well-established techniques existed for anchoring the DNA to a glass plate. Having demonstrated that light-directed synthesis of peptides using photolithographic masking technology was possible, and knowing that all the pieces for doing a parallel DNA synthesis were within reach, Fodor was eager to shift his attention entirely to developing the gene chip. In a fax of May 15, 1990 to Stryer, Fodor outlined the reasons for his convictions that it was time to devote full concentration on the gene chip. The upshot of this was to spin off the gene chip project as its own company, Affymetrix (for Affinity Matrix).

Fodor's prototype of the light-directed parallel peptide synthesis array, the fluorescence scanner and computer system for keeping track of each spot on the chip and quantifying the ratio of tagged DNA matches as color spot ratios in a computer graphic output, the photomasks and appropriate photochemicals for constructing the arrays were all designed in the heady "think tank" environment of Affymax. The discussions in the Affymax scientific board meetings allowed Fodor and his colleagues to draw upon the knowledge and vision of some of the leading academic biochemists and chemists of the day from several universities, including Stanford, Berkeley, Cal Tech, and Lawrence Livermore Labs. This back-and-forth flow of information between academic researchers and the efforts to launch the company had very much the style and spirit of a Silicon Valley startup.

The distributed networked character of innovation in the Silicon Valley is exemplified by Fodor's original prototype system for using photolithography to design a peptide microarray. As a way to construct chemical diversity, Stryer suggested laying down a grid of parallel stripes – each stripe with a different compound – one compound at a time, then repeating the procedure with stripes laid down perpendicularly on the grid, one at a time. In order to see how to work it out using photolithography, Fodor contacted Fabian Pease, professor of electrical engineering at Stanford, specializing in electron beam lithographic mask fabrication. Pease, a Ph.D. from Cambridge University, had been an assistant professor at UC Berkeley briefly before moving, in 1967, to Bell Labs, where he first worked on digital television and later led a group that developed the processes for electron beam lithographic mask manufacture and demonstrated a pioneering LSI circuit built with electron-beam lithography. Pease had been at Stanford since 1978. Fodor and Stryer persuaded Pease to join Affymax as a consultant on their project, and he and Fodor spent a lot of time discussing technical aspects of lithography needed to build the microarray. Pease took Fodor around to various warehouses in Silicon Valley to acquire old lithography instruments needed for building the prototype peptide array. By May, 1990 with periodic input from Pease, Fodor had a working semi-automated lithography instrument that would do binary combinatorial peptide syntheses. Pease maintained his connection to Fodor after the launch of Affymetrix in 1992. In 1993–94, for instance, he took a sabbatical from Stanford to work on the DNA microarray. Pease has been co-inventor along with Fodor and Stryer on several key Affymetrix patents, and he has continued to maintain a consulting relationship with Affymetrix [Note F].

A similar story of Silicon Valley networking led to the design of the first microarray scanner and reader. Through the network of contacts of the Affymax scientific board, Fodor got in touch with Peter Fiekowsky to assist him in the development of a system for detecting and imaging the fluorescently labeled markers of polymer sequences on the peptide array. Fiekowsky had received his BS degree in physics from MIT in 1977. Following graduation, he moved to Silicon Valley to work at NASA and moved to Fairchild's artificial intelligence lab in 1983, where he worked on image analysis in the semiconductor industry. A year later, in 1984, Fiekowsky founded Automated Visual Inspection. The work he and Fodor did on the array project led to two of the 23 patents Fiekowsky holds in image-processing techniques ranging from semiconductor and flat panel inspection to medical x-rays and gene chips 1314.

James L. Winkler's involvement with Fodor and the core technologies in the launch of Affymetrix provides another typical example of the wide range of talents and the veritable gene pool of innovators who circulate through startups in Silicon Valley. As Fodor recalled in an interview, "Winkler was one of these guys who was just brilliant, did not have any formal education, but could build anything. He could take a blank circuit board and by the end of the day have something he could plug into the back of the computer to run an external piece of equipment." [Interview with Stephen Fodor, August 2004] One of Winkler's first contributions was the design and implementation of the method and devices for flowing reagents through block channels on the glass microarray substrate to form the stripes of different peptides in combination with the light-directed method of coupling and decoupling. After each stripe was laid down, the substrate was shifted by a rotating stage, and the process repeated to form arrays of polymers on the substrate 14. This was just the first of what would become 31 patents on different aspects of gene chip production and photolithographic mask design, including a set of computer tools for selecting probes and designing the layout of an array of DNA or other polymers and using chip design files to design and/or generate lithographic masks 16.

The guidance that Affymetrix received in its nascent years from consultation arrangements with academics and other local Silicon Valley experts was crucial to the advance of gene chips and related systems. Research in several domains had been going on for years in university and government research projects that provided fertile sources of ideas and techniques for developing the complex technology of the DNA microarray. In fact, university scientists appeared several dozen times on granted Affymetrix patents, although some of these can be accounted for by university faculty who had been hired into the company [Note G]. In Table 1 we present the results of our scan of the patent data for academic collaborations with Affymetrix [Note H].

We believe that these university collaborators provided enabling expertise to Affymetrix, without these ongoing consultations the development of the microarray would have taken much longer. Federal funding has been particularly important in the development of microarrays. On the one hand, as we have seen, federal funding for extra-university-based industrial research and development provided the capital to launch the cluster of innovative technologies directly connected with the GeneChip^® at Affymetrix; and as we shall show in our case studies further on, federal funding was crucial for the take off of some competitor technologies in the microarray field. But the work at Affymetrix and other firms in the microarray field was heavily dependent on knowledge and expertise that had accumulated in several academic disciplines, including biochemistry, genetics, electrical engineering, and computer science as a result of at least two decades of federal funding from the NSF, NIH, DOE, and programs such as the Human Genome Initiative, particularly at Bay Area universities, Stanford, UC Berkeley, and UCSF. In terms of the infrastructure of innovation discussed above in our introduction, "knowledge spillovers" from these federally supported academic research programs provided important resources to the nascent field of microarrays. Federal funding of extra-university research and development by industry provided the stimulus for drawing those resources into an accumulating ensemble of innovations that gave rise to a major new technology and several new lines of research. Government support, particularly to nearby universities, lowered the cost of development through the cultivation of experts who played a pivotal role in the creation of the GeneChip^®.

The development of combinatorial chemistry, microarrays, and the GeneChip^® at Affymax and Affymetrix and other Bay Area companies calls attention to an important but often overlooked feature of the development of high technology regions: namely, the role of federal funding for research and development in companies that transforms the academic research environment while launching new industrial sectors. Most discussions of federal funding for research concentrate on the role of federal funding in driving academic research. But as our analysis of the rise of Affymetrix demonstrates, federal funding has also been crucial in stimulating the other side of the equation in the symbiosis of Silicon Valley and research universities such as Stanford: namely, in the formation of the startup companies and collaborations with large established companies in the development of new innovative technology. We frequently point to the massively central role of the federal government in funding academic research, but it is also the case that in Silicon Valley the government has played and continues to play a large and absolutely vital role in funding new industrial development. This point has been made frequently about the role of defense contracting in support of early developments in the electronics and semiconductor industries during the 1950s–70s. But federal funding has also been a major factor in the development of biotech, materials science, and several related industries from the 1990s to the present.

Table 2 and Figure 2 illustrate the significant contribution of federal funding of both university and industry R&D in California for the period of the 1990s to 2002.

During the decade of the 1990s through the early 2000s California ranked number one among states receiving federal funding for research. During this period the average annual federal obligation to California R&D in industry was approximately $6.96 billion, while support of university-based R&D averaged approximately $3.3 billion. Although the trend line points to a decreasing amount of federal spending for industry R&D in the later years of the period with an encouraging increase to universities, the fact is that federal support of California R&D nearly tripled support for university-based research. Of course, a sizeable portion, typically exceeding 50% of the total Federal R&D in California is directed toward the defense industries. But even allowing for defense spending and not taking into consideration that some biotech research is funded by the DOD, the amount of non-defense related federal funding to industry in California exceeds federal support for academic research by a considerable margin – typically by a factor of two Particularly important for the companies like Affymetrix, Symyx and other startups in the microarray field we will discuss below is the number of Small Business Innovation Research Program (SBIR) awards and Small Business Technology Transfer Program (STTR) awards going to California [Note I]. Throughout this period California has averaged around 1,000 SBIR and STTR awards, ranking first in both categories of awards, with Massachusetts typically ranking second. Moreover, during this period California has received 184 awards from the Advanced Technology Program, a program that sponsors startup companies having a university-based collaboration or academic PI. From the perspective of Silicon Valley, the 1990s were the best of times. If we compare all SBIR and STTR awards received by firms in the Bay Area zip codes that constitute Silicon Valley versus all California awards, the Bay Area has averaged 33 percent (an average of $62 million per year) of the awards with a high of 39 percent in 1993 and a low of 25 percent in 2002. Several Bay Area companies, such as Affymetrix, have been the recipients of multiple SBIR/STTR/ATP awards. Figure 3 provides an overview of SBIR and STTR awards specifically to the Bay Area.

When we consider that 58 percent of the federal funding for R&D to universities in California has gone toward funding innovation in the life sciences (see Table 2), the importance of the NIH and the Human Genome Project for the explosion of biotech firms in Silicon Valley becomes evident. Federal funding has also been significant in sustaining an entrepreneurial academic environment at Stanford and other Bay Area universities that have participated in numerous waves of technological innovation within the Silicon Valley through the students they train and the faculty engaged in research and consulting as well as in working with their university technology licensing offices to disclose, patent, and license inventions. As we have shown in another study, Stanford's openness to (in former Stanford Dean of Engineering, Jim Gibbons' phrase) "reverse engineering," the enhancement of new research directions through absorption of technological directions emerging in the Silicon Valley as key to its entrepreneurial culture, is one of the pillars of its success. Stanford receives approximately $500 million in federally funded research grants annually. Berkeley and UCSF are also in the top 20 research universities receiving federal support. As we now see, federal funding is also deeply involved in stimulating and sustaining the reverse engineering essential to this co-evolution of Bay Area research universities and the Silicon Valley.

Affymetrix was well positioned to take advantage of the flows of information from both the academic and biotech communities within Silicon Valley to acquire funding and intellectual resources necessary for assembling the pool of ideas, inventions, and know-how behind the microarray. Local university help and federal funding were essential to Affymetrix's push to begin developing the GeneChip^® in 1990. Encouraged by Stryer and increasingly confident about the success of the GeneChip^®, Fodor sought to capitalize on the wave of funding for technologies associated with the NIH's goal to uncover and exploit genetic information. The Human Genome Project had been launched a few months earlier and the NIH was soliciting proposals for the development of technology in support of genomics. Ron Davis from the Stanford Biochemistry Department, together with David Botstein from Genetics, were developing technology for the human genome sequencing effort at that time. Stryer invited Davis over to Affymax to discuss the use of the gene chip technology to perform genetic sequencing by hybridization. Paul Berg, who was on the Affymax scientific board, was also interested in the technology, and he attended the meeting. Both Davis and Berg immediately saw the potential of the technology, and Davis was excited enough about what he saw to propose a collaboration with Fodor to apply for NIH funding to support the development of the gene chip. The NIH panel for sequencing technology for the Human Genome Project directed by Leroy Hood was meeting across the bay in Walnut Creek, CA in the spring of 1991, and Paul Berg arranged for Fodor to be invited to present on the peptide and DNA chip project. James Watson and a blue ribbon panel of genome scientists were in attendance, and when the meeting concluded, Fodor and Davis were encouraged to apply for funding. In September 1992 the first of several grants to Affymetrix was awarded with Stephen Fodor as PI. Co-PIs on the project were Ron Davis from Stanford and Ronald Lipschutz from Daniel H. Wagner Associates, a mathematics firm that contributed expertise on improving algorithms for sequence analysis [Note J]. The initial NIH grant, funded from 1992–95, was for $2.5 million, and together with a Phase I Small Business Innovation Research (SBIR) grant from the Department of Energy (one of several SBIR grants the company has received) for $500,000 awarded in 1992, Fodor was able to demonstrate proof of the concept of using large arrays of DNA probes in genetic analysis. A Phase II grant was awarded to assist Affymetrix in moving the technology towards commercialization. Scientists at Affymetrix also received several grants from the National Institutes of Health. For example, Fodor was principal investigator on a second round of NIH funding in 1995 for a three-year $5.5 million NIH, grant from 1995–97. One component of this grant addressed the development of chip-based sequencing, re-sequencing, sequence checking and physical, genetic, and functional mapping. A technology development component addressed the production of chips and the development of instrumentation and software specific to the chip applications.

Affymetrix's largest government award in the startup phase of the company came from the Advanced Technology Program (ATP) of the National Institute of Standards and Technology (NIST) in the Tools for DNA Diagnostics Focused Program competition in 1994. In its reports documenting the successes of its programs, the ATP lists Affymetrix as one of its banner projects 17. The ATP program was started in 1990 to stimulate new science-based research ventures and to encourage joint ventures among universities, industry, research organizations, and consortia of companies. A consortium established by Affymetrix was awarded a $31.5 million, five-year grant in 1994 to develop miniaturized DNA diagnostic systems. Under this grant, Affymetrix directly received $21.5 million, some of which was used to fund activities at a number of collaborating institutions as subcontractors to the project. As part of this grant, Affymetrix and its partner Molecular Dynamics collaborated with researchers at the California Institute of Technology, Lawrence Livermore National Laboratory, Stanford University, the University of California at Berkeley, and the University of Washington to develop the next generation of diagnostic devices to capitalize on the advances of the Human Genome Project. After developing its core chemical synthesis technology while still funded under the ATP and SBIR grants, Affymetrix entered into agreements with OncorMed to collaborate in development of clinical validation of genetic testing services utilizing the GeneChip^® for analysis of genes associated with cancer; and under a separate distribution and instrumentation alliance between Affymetrix and Hewlett-Packard, Hewlett-Packard began developing and supplying a next-generation scanner to read the GeneChip^® in 1996. The Advanced Technology Program was particularly enthusiastic about the ways in which Affymetrix accelerated the diffusion of its technology through alliances and collaborations with the Genetics Institute, Roche Molecular Systems, Incyte Pharmaceuticals, and Glaxo Wellcome in order to continue raising capital for expanding its own internal R&D 18. Table 3 tracks federal funding that Affymetrix received over a ten year period [Note K].

Two themes emerge from the way the government funded Affymetrix: the wide range of government organizations that provided the funding, and the variety of federally funded research projects at Affymetrix. The diversity of agencies that saw benefits to the GeneChip^® is quite apparent: the Department of Energy, NASA, and several organizations within the National Institutes of Health funded Affymetrix over the eleven-year period studied. Later affirmed by the breadth of research applications the DNA chips found, this broad set of government health organizations, such as the National Cancer Institute, the National Institute of Allergy and Infectious Diseases, and the National Institutes of Neurological Disorders and Stroke, provided early testimony to the GeneChip^®'s widespread applicability.

These kinds of collaborative research efforts were a prerequisite to acquiring federal funding to launch the company, and they have continued ever since to be a deliberate core strategy of Affymetrix, carried over from Affymax, to maintain simultaneously within the firm an entrepreneurial as well as an academic environment. The firm's goal was to attract preeminent researchers and convince them that the company was creating cutting-edge technology. Steve Fodor was persuaded to leave his postdoctoral research position at UC Berkeley – despite his initial lack of interest in leaving academia – by the possibility of continuing to work with some the field's brightest academics as well as having in-house funding to do research. The freedom to seek outside grants to pursue research peripheral to the company's core strategies was also considered an important tool in attracting high-quality people to the project. Affymetrix has been able to attract staff who continue to keep their academic contacts through participation in grant proposals, and who have the freedom to pursue ideas to which they have dedicated their careers, while gradually migrating to a commercial environment where more tangible products can be generated. The exercise of building a consortium of other companies to work together under the ATP project, for example, fed a very collegial environment where researchers worked hard with the best people in their field around the world, pushing these technologies to a stage at which they could be commercialized successfully.

The 1991 paper in Science on parallel chemical synthesis using microarrays inaugurated the field of combinatorial chemistry, and it may indeed be one of the key events in the genomics revolution. By 1999 articles in Science among many other scientific journals were celebrating the widespread use of microarrays and the way they had transformed genomics 19. People who never thought they would do large-scale gene studies suddenly were eager to try their hand at monitoring thousands of genes at once. The National Institutes of Health (NIH) heavily supported this trend, funding its own microarray studies and providing grants to institutions to buy the technology. This generous support of studies using microarrays generated a flood of data that traditional journals found hard to accommodate and digital databases didn't yet know how to handle. The NIH funded workshops to spread the technology. A Cold Spring Harbor Laboratory workshop on microarrays led by Pat Brown from Stanford in 1999, for instance, was the most over subscribed laboratory course on record in the history of Cold Spring Harbor programs. The new course was not even advertised, yet eight times as many people signed up as could be accepted. Sixteen people paid $1955 each to learn how to build and use a machine for genetics research. For another $30,000, four actually took the machine home.

Research and development of microarrays was the hot new field in the 1990s. Although their approach was distinctive in focusing on in situ synthesis of DNA libraries on a chip, Affymetrix was not alone in the microarray field. About the same time the Affymetrix group was developing the GeneChip^®, several academic teams were developing alternative microarray systems [Note L] 20. Of particular importance were spotted microarrays developed at Stanford by Pat Brown, Dari Shalon, Stephen J. Smith, Mark Schena and Ron Davis. The Stanford system was a contact array that used two-color fluorescence hybridization. On the heels of this system was a non-contact array developed by Leroy Hood at Cal Tech that adapted the technology for ink-jet printers to micro spot solutions of nucleotide reagents printed on a glass substrate 212223.

The spotted microarrays were extensions of methods that had been in use in genome analysis and molecular biology for two decades, going back to Edwin Southern's introduction of the Southern Blot 24. Another forerunner for all the microarray work, including the work of Fodor et al., were the methods for locating the position of specific sequences in chromosomes through fluorescence in situ hybridization (FISH), which allowed cell nuclei and chromosomes to be fixed to glass microscope slides as solid support. Ron Davis had contributed to those early methods for identifying genes, and the same technique was used to fix DNA to slides as solid support for his later microarray work 25. The technique of using ordered arrays of DNA at the core of microarray techniques also grew out of earlier work. Of special importance was the dot-blot method introduced in 1979 by Fotis Kafatos, et al., in which hybridizations were carried out in parallel and fluorescent signals representing hybridization were measured with an imaging method 26. The procedures for constructing these arrays were manual and the spots, as in the Southern Blot method, were deposited on various types of porous filters. While effective, these early spotting methods on porous materials were not suitable for the large-scale genome analyses that took off in the 1990s: it was not possible, for instance, to reduce the size of the spots beyond certain limits, or to control their size and shape on a porous membrane. The large scale automation of these dot-blot procedures was undertaken by Hans Lehrach and his co-workers at the Berlin Max-Planck-Institute for Molecular Biology in 1994. Lehrach's group developed laboratory robotic systems for picking and spotting clones onto filters 27. This move toward large-scale automation with robots coupled with the replacement of the porous materials used in dot-blots with impermeable supports, such as glass or silicon, were key steps in the development of the spotted microarray systems. Non-porous surfaces permitted the use of very small sample volumes and high sample concentrations of spots. Over the next few years during the early 1990s technical advances made it possible to generate arrays with very high densities of DNA spots, allowing for tens of thousands of genes to be represented in areas smaller than standard glass microscope slides. These changes to the macroscopic format of filter based arrays resulted in the miniaturized "biochip" format of the microarray that has brought about a fundamental revolution in biological analysis. By effectively making it possible to represent the entire genome of an organism on a single biochip, researchers are able to study the expression of all the genes of a particular organism at once.

The spotted microarray developed in Pat Brown's lab consisted of two principle pieces of hardware; the arrayer and scanner 28. The arrayer was a variation of the standard "pick-and-place" XYZ-axis gantry robot common to many large university molecular biology laboratories. Glass slides coated with a poly-lysine surface were placed on a platter. The robot picks up pre-synthesized single strand or double stranded DNA samples from a 384 well microtitre plate by placing a specially designed cluster of spring-loaded printing tips into adjacent wells of the source plate, each tip filling with approximately 1 micro liter of DNA solution. The DNA samples are, in most cases, labeled by incorporating fluorescently tagged nucleotides. The cone-shaped printing tips in Brown's original system were stainless steel with manually sharpened points and a slit up the center for holding the DNA solution. They operated on the same principle as a quill pen; liquid was drawn up by capillary action and deposited when the tip made contact with the slide surface. The printing tips are tapped leaving a small (less than 0.5 nano liters) drop at identical positions on each slide. With the spacing between tips deployed in the microarrayer the entire human genome could be spotted onto a standard 1-inch by 3-inch laboratory slide.

After hybridization a fluorescent image of the array is acquired by a laser scanning confocal microscope. The scanner has a laser (or lasers) producing light at the appropriate wavelength for the excitation spectra of the two dyes (red and green) being used. The light passes through the microscope objective and illuminates a single point on the slide. The emitted light gathered by the objective is filtered to remove the excitation beam, passed through a pinhole (removing noise), and finally quantified in a photomultiplier tube. The relative amount of fluorescence is measured for each spot on the array using software Brown's team developed for segmenting the images into boxes and determining the average fluorescence for each box. The advantage of using fluorescent signals is that they do not disperse, and accordingly allow for very dense array spacing. Also a significant advantage of using two or more differently labeled probes targeted to the same spot in this system is that each can be detected separately. In this way, two-color hybridization detection allows for a direct quantitative comparison of the abundance of specific sequences between two probe mixtures that are hybridized competitively to a single array.

Brown, Shalon, and Smith 29, and Davis and Schena 30 have argued that spotted microarrays have several advantages over the in situ chips designed by Affymetrix and Edwin Southerland. As we have seen in the case of GeneChip^® design, in situ synthesis methods work with oligonucleotides, libraries of nucleic acid sequences of between 2–25 base pairs. On a GeneChip^® a given gene might be represented by 15–20 different 25-mer oligonucleotides that serve as unique sequence-specific detectors. To be effective, the Affymetrix arrays require gene sequence information for specifying the de novo synthesis of the oligomers on the array. Spotted microarrays by contrast represent genes by single DNA fragments greater than several hundred base pairs in length, and virtually any length or origin. Moreover, spotted arrays do not require prior sequence knowledge but can be produced from both known and unknown cDNA and PCR fragments. Spotted microarrays, it is argued, are more flexible and more easily adaptable to a variety of research problems in genomics. Also to the point, spotted microarrays are inexpensive by comparison to Affymetrix chips 31. Indeed, microarrayers based on the Brown-Shalon design could basically be constructed in-house by most major university research labs at a complete cost (in 1999) of around $60,000 32. Brown in fact has been so committed to the low cost production of microarrayers and an open source approach as a means to expedite the production of knowledge in genomics that he posted on his Stanford website all the details of manufacture for his microarray system, including all the software updates for operation of the scanning system, details on manufacturing and servicing the printing tips, and other fine points of the system.

To study the adoption of both in situ and spotted microarray technologies, we considered the first academic articles either reporting studies based on using DNA chips or simply discussing DNA microarrays. We focused on pre-1999 studies because the DNA chip-based research began to take off in 1999. These articles broke down into four main types: results of microarray studies, overviews of how to use gene chips, technology forecasts, and descriptions of new or otherwise improved DNA chips. As an indication of what types of studies were represented in the early publications about gene microarrays, we present Table 4:

Of the early articles we studied, by far the majority reported on the results of experiments using DNA chips. Most of these studies aimed to uncover significant genetic information in areas of existing interest, such as cancer and cardiovascular disease; to understand the role of genes already identified as being important to particular diseases; or to attempt wide-scale gene expression monitoring of organisms whose genomes were already heavily studied, such as the Arabidopsis plants and Saccharomyces yeasts. By addressing several different research communities, these initial studies served to broadcast the potential of the new microarray technology. While these studies were excellent advertisements for the technology, they also opened up promising avenues of inquiry, helping the technology establish itself in a variety of research areas.

Getting involved with the technology early on was not simply a matter of desire; generally, early authors had some affiliation with Affymetrix. In part because of their longstanding relationships and ongoing collaborations with Affymetrix and due to their internal microarray development efforts (largely arising from their collaboration on the Human Genome Project), Stanford and the NIH also possessed a great deal of in-house expertise in using microarrays, which allowed them to assist researchers from other organizations in using the technology. In fact, when we tabulated the affiliations of scientists appearing on the pre-99 microarray studies, we found that Stanford, NIH, and Affymetrix appeared most often. Table 5 lists the most frequently occurring author affiliations within the set of 130 early articles on microarrays [Note M]:

Perhaps unsurprisingly, the first organizations to publish studies based on research using DNA chips were often those that had the strongest links to gene chip manufacturers. In fact, the second most common organization to appear as an author affiliation among the 130 studies we surveyed published prior to 1999 was Affymetrix. The top organization was Stanford, which had collaborated extensively with Affymetrix in the development of the GeneChip^®, was developing its on pin-based arrayers, and possessed a great deal of in-house expertise in using the gene chips. The NIH's massive network of intramural research and its strong links (including research collaborations) to Stanford and Affymetrix made it third.

It was not simply a matter of being involved with the development of various microarray systems that led to publication, later research collaborations with and between these expert organizations also coincided with earlier and more frequent publication. For example, 28% of articles with a Stanford author also had an Affymetrix scientist and researchers from Stanford spin-out Synteni (whose technology was largely based on the Dari Shalon and Pat Brown system) appeared on 20% of the Stanford publications.

The top three organizations listed above – Stanford, Affymetrix, and the NIH – were the major "hubs" (or highly connected points) in co-authorship networks for the 130 studies we surveyed. To study the network of collaborations during this early phase of research using gene chips, we used an analysis tool that graphically places organizations according to their co-authorships with other organizations [Note N]. For example, in Figure 4, Affymetrix (large green node) co-authored with Princeton, but Princeton did not co-author with NIH (large blue node), so Princeton is near Affymetrix but distant from NIH. Furthermore, although Affymetrix and NIH co-authored papers together, they also co-authored papers with several organizations that did not co-author with both organizations, thus Affymetrix and NIH are pulled some distance apart (as opposed to NIH and Stanford, the large red node, which share more institutional co-authors) [Note O].

There were many organizations represented in the first 130 articles dealing with DNA microarrays, but Stanford, Affymetrix, and the NIH emerge as major nodes in this network. Often, other organizations would partner with one or more of these major players and then go on to collaborate with organizations previously outside the network. The heavy overlap of collaborations indicates that this was a fairly tight-knit research community. Several organizations in the center and upper right of the map collaborated with at least two of the three major players. Interestingly, many of the initial participants in microarray based research were also involved in the Human Genome Project.

Institutions that were the first to publish microarray studies and that collaborated with DNA microarray makers were also the best able to attract federal funding for microarray based research [Note P]. Organizations that appeared in Table 5 as having been the first to publish studies based on gene chip research tended to be those that received the most federal grants for DNA microarray research over the period 1993–2004 (shown in Table 6) [Note Q]. This particular phenomenon in the academic setting of being first to collaborate with Affymetrix, subsequently being first to publish DNA microarray based studies, and in turn receiving more federal funding is roughly analogous to the type of positive feedback loop that economists have used to describe how initially successful high technology firms become increasingly entrenched within their industries.

DNA microarray makers such as Affymetrix have been hubs for an expanding network of companies and technologies across the spectrum of technologies fueling contemporary biotech, gene-based medical therapies, and areas of materials science. These companies have drawn heavily upon academic researchers as consultants and scientific advisory board members, and they have collaborated with academic researchers in sponsoring postdoctoral work and a variety of research projects funded by the NIH, NSF, DOE, and other federal agencies. The academic researchers involved have only in rare cases relinquished their university positions to move into industry. While some of these individuals, such as Schultz, Berg, Stryer, Ron Davis, Mark Davis, and others have been involved in numerous startups, they have returned to their universities (Stanford and UC Berkeley) where they have continued to develop graduate programs that incorporate these new innovations. Other Stanford faculty, such as Fabian Pease and Calvin Quate, have continued as advisors and collaborators in shaping new generations of microarray and sequencing technologies at Affymetrix. Through these technologies and the academic researchers who have participated in developing them, research programs at Stanford and other universities in a variety of different disciplines have taken new shape and direction.

In order to trace the widespread impact of microarrays on the academic research environment, Table 7 presents a chronological overview of the interest in microarrays and gene chips by several disciplines as indicated by citations to the first 130 articles published based on microarray research. (For the totals from nearly all fields citing microarray research, see Appendix A) [Note R]. The data show that interest in DNA chips and microarrays more generally was manifest in a variety of disciplines. As a new, promising, but unstable and unproven technology, microarrays were attractive as a platform that could be improved upon by many different fields. In an era when researchers were motivated to find new ways to interpret the massive amounts of data being generated by the Human Genome Initiative, researchers in just about every field of biomedicine were looking for novel high-throughput techniques to refine genetic analysis and develop tools for rapidly interpreting gene expression data. In many of the new areas, the microarray and gene chip were tools for advancing a program of "molecularizing" established disciplines. But this could not be accomplished by simply plugging in a microarray and reading off the results. New tools and even modifications of the gene chip itself had to be developed in order to assimilate the microarray to the research objectives of these several fields. Multidisciplinary teams of researchers and collaboration between academic researchers and their industry partners proved essential to advancing the technology. The demand for alternatives greatly expanded the market for these research tools and, as we show below, created opportunities for other firms to enter the market.