41J Blog

Early promo shot from the old Solexa site. The 90s rocked!

In this post I describe non-technical aspects of the development of the Solexa sequencing platform. For those who don’t know, Solexa is the company that was acquired by Illumina. They developed (acquiring the necessary IP as required) a DNA sequencing platform several orders of magnitude faster and cheaper than what was previously available. The core chemistry developed at Solexa returns broadly similar to that currently used by Illumina, and which dominates the DNA sequencing market. If we decided to implement GATTACA today, you’d probably use Illumina machines.

Solexa are now long gone of course, the solexa.com now proudly boasts an exciting deep clean make up remover. Old versions are available via the waybackmachine.

Kind of a shame Illumina didn’t keep the domain…

It’s interesting to look at how things played out financially. And what it cost to get to each milestone. In this post I’ve pulled some data out of their filings with companies house, and a SEC filing which describes their technological progress.

Briefly the following table describes the funding I’ve seen noted in the accounts and press releases (this maybe wrong and I’d welcome corrections):

It’s instructive to plot this against certain milestones. These are available in a Solexa SEC filing.  You can see that they’d raised 34.4MUSD in total prior to sequencing anything at all. Getting the platform up and running, took time. The sequencing milestone is 9 months after they acquired the Manteia cluster amplification IP which proved critical. Things moved rapidly after this point.

The total amount raised tells us something about how much it cost to develop the technology. It doesn’t however tell us about the valuation of the platform at each round. If you just plot the amount raised (rather than the total) you get the following:

The fact that the raise in 2004 was lower than the previous value makes me suspect that it could have been a down round? Based on the numbers calculated in the section below I don’t think so. The valuation possibly looked like this:

The interesting thing about this is that the valuation prior to a proof of concept looks like it was in the ~30->60MUSD range.  After getting sequence data the valuation must have increased rapidly. It feels like the acquisition price was quite low, and that it could have been single digit multiples of the last round valuation. So, those are my thoughts on Solexa’s valuation and fund raising, I will try and rework this and check/fix the calculations when I can. Below I detail some the background/sources and how I arrived at some of the numbers above.

Business Development Background and Sources

Progress in data generation at Solexa from 1.

The company was incorporated in 1998 as “Intercede 1356 Limited”. As a UK based company all its prior accounts are a matter of public record. However you’ll now have to search under “Illumina Cambridge Limited” (the business unit Solexa became after acquisition).

Their initial funding came from Abingworth Bioventures II. The first set of accounts filed shows 300KGBP in cash and records a 300KGBP loss (it records transactions to the end of 1999). The company commenced activities on the 12 of October 1998. And I would expect this raise to have been around that point.

The year 2000 accounts are a little more explicit, and contain the following statement:

REVIEW OF THE BUSINESS AND FUTURE DEVELOPMENTS

The Company has been established to undertake research and put the results of such research to commercial use.

The Company raised a further £1.5 million during the year by way of a convertible loan from Abingworth Bioventures II SICAV (see note 8 to the accounts). This has been used to establish the Company in its own premises and to build the management and scientific teams. Good progress has been made in the development of the Company’s proprietary single molecule technology.

EVENTS SINCE THE BALANCE SHEET DATE

Subsequent to the year end the company raise £100,000 by way of a convertible loan from the University of Cambridge.

On 20 September 2001, the capital element of the above convertible loans of £1.6 million from Abingworth Bioventures II SICAV and University of Cambridge were converted into 800,000 ordinary shares of 0.25p each at £2 per share. In addition £4,104 of accrued interest on the University of Cambridge loan was converted into 2,052 shares at the same rate.

Also on 20 September 2001, the company raised £12,000,000 (before expenses) by the issue of 4,000,000 ‘A’ ordinary shares of 0.25p each at £3 per share.

So by my reading there was an initial investment of 600KGBP in 1999. Sometime in 2000 they got an additional 1.5MGBP from Abingworth. Then in 2001 an additional investment of 13.6MGBP. It’s also an interesting technical note that they describe themselves as a single-molecule sequencing company.

Going through the accounts the next interesting event is the acquisition of the Manteia IP:

On 25 March 2004, Solexa Ltd and Lynx Therapeutics Inc. jointly acquired from Manteia SA the rights to propriety technology assets for DNA colony generation. Solexa intends to use the Intellectual Property, in conjunction with its existing technology for the comprehensive and economical analysis of individual genomes.

This was the point that they transitioned from being a single molecule company to a cluster (or as they called it at the time colony) based sequencing platform. For those familiar with the technical aspects of the system this, is a significant change in direction for the company.

Prior to 2004 they were burning about 3 million a year it seems. In 2004 they were burning about 5 million (GBP). By my reading of the accounts they had about 5MGBP left in the bank, that’s after raising what looks like 8MGBP:

On 30 July 2004 the Company issued 4,166,666 ‘B’ preferred shares with an aggregate nominal value of £10,471 for gross proceeds of £7,500,00. On 18 October 2004 the Company issued a further 277,778 ‘B’ preferred shares with an aggregate nominal value of £694 for gross proceeds of £500,000.

That was probably raised from Amadeus (as noted in this press release), in fact the Amadeus press release notes 14.4MUSD which is almost exactly 8MGBP at that dates exchange rate. The press release also notes that in total Solexa had raised 40MUSD to date. The exchange rate in 2004 was almost 2 dollars to the pound, so this approximately matches the 22.2MGBP if you total up what I found in the accounts.

In March 2005 they were acquired by Lynx Therapeutics, Inc.. While structured as an acquisition, it was more like a merger (or Solexa acquiring Lynx) Lynx changed its name to Solexa, Inc continuing the Solexa brand. The deal was possibly largely done to give Solexa a US stock listing. I believe most of the tech development was still done by the Solexa guys in Cambridge.

Because of the merger, the UK accounts are less informative from this point on. But we can see that in 2005 they burnt 9MGBP. The accounts state they have 8MGBP left in the bank. And 2006 they appear to have burnt about 13.5MGBP. And in 2007 they were acquired by Illumina for what was widely reported as being 600MUSD.

Other sources of information help fill in the gaps. A press release from July 12 2005 contains the following:

Solexa Completes $24 Million Private Equity Financing

HAYWARD, Calif.–(BUSINESS WIRE)–July 12, 2005–Solexa, Inc. (NASDAQ:SLXA) today announced that it has completed a private equity placement for approximately $24 million following stockholder approval of the financing at the Annual Meeting of Stockholders held on July 7, 2005. The financing represented the second and final closing of the $32.5 million private equity placement that was announced on April 21, 2005. SG Cowen & Co., LLC served as the exclusive placement agent for the transaction.

“This financing demonstrates our investors’ confidence in our ability to execute on our business plan to develop and market our next-generation sequencing systems based on Sequencing-by-Synthesis (SBS) and Cluster molecular arrays,” said John West, Solexa’s chief executive officer. “In the coming months, as we move closer to product launch, we expect to be able to announce additional experimental results demonstrating the performance of our platform in high-end genetic applications.”

Under terms of the financing, the second closing included the sale of approximately 6.0 million shares of common stock at $4.00 per share and issuance of warrants to purchase up to approximately 3.0 million shares of common stock at an exercise price of $5.00 per share. The first closing of the private equity placement, completed April 25, 2005, generated proceeds of approximately $8.5 million from the sale of approximately 2.1 million shares of common stock and approximately 1.1 million warrants. As previously announced, Solexa’s prior venture capital investors Abingworth Management Limited, Amadeus Capital Partners Limited, Oxford Bioscience Partners and SV Life Sciences invested a total of approximately $10.8 million in the financing at the second closing.

Stockholders at the company’s annual meeting also approved all other items included in the company’s 2005 Proxy Statement. Among the measures were the election of seven nominees to serve on the Solexa board of directors for the ensuing year, including three affiliated with the company’s venture capital investors and one with ValueAct Capital, the lead investor in the private equity financing. Other approved proposals included adoption of the company’s 2005 Equity Incentive Plan.

Their early press release contains the same information:

HAYWARD, Calif.–(BUSINESS WIRE)–April 22, 2005–Solexa, Inc. (Nasdaq:SLXA) today announced that it has entered into a definitive agreement for a $32.5 million private sale of common stock and warrants for the purchase of common stock with a group of leading institutional investors in the health care sector. The transaction is led by ValueAct Capital. Solexa also announced that G. Mason Morfit, CFA, a partner of ValueAct Capital, has been appointed to its board of directors, bringing board membership to eight.

Under terms of the financing, Solexa will sell approximately 8.1 million shares of common stock at $4.00 per share and will issue warrants to purchase approximately 4.1 million shares of common stock at an exercise price of $5.00 per share. Approximately 2.1 million shares of common stock and approximately 1.1 million warrants will be issued in a closing expected on or about April 25, 2005, and the balance of approximately 6.0 million shares of common stock and warrants to purchase approximately 3.0 million shares of common stock will be issued on the same terms in a second closing subject to stockholder approval at the 2005 annual meeting. Solexa’s intended use of proceeds includes the development and launch of its first-generation Sequencing-by-Synthesis (SBS) molecular array platform for genetic analysis and repayment of its loan from Silicon Valley Bank. All of Solexa’s previous venture capital investors, including funds affiliated with Abingworth Management Limited, Amadeus Capital Partners Limited, Oxford Bioscience Partners and SV Life Sciences, will be investing a total of approximately $10.8 million in the financing at the second closing of the financing and have entered into agreements with Solexa to vote in favor of the financing at the 2005 annual meeting. SG Cowen & Co., LLC served as the exclusive placement agent for the transaction.

A press release from 11/21/05 contains the following statement:

Solexa Announces Agreement for $65 Million Private Placement

HAYWARD, Calif. and CAMBRIDGE, England–(BUSINESS WIRE)–Nov. 21, 2005–Solexa, Inc. (Nasdaq:SLXA) today announced that it has entered into a definitive agreement with a group of institutional investors to raise approximately $65 million from the private sale of common stock and warrants for the purchase of common stock. This financing will result in net proceeds to Solexa of approximately $61 million after deduction of offering expenses.

Under the terms of the financing, Solexa will sell 10.0 million shares of common stock at $6.50 per share and will issue warrants to purchase approximately 3.5 million shares of common stock at an exercise price of $7.50 per share. Approximately 3.9 million shares of common stock and approximately 1.3 million warrants will be issued in a closing expected on or about November 22, 2005, and the balance of approximately 6.1 million shares of common stock and warrants to purchase approximately 2.2 million shares of common stock will be issued on the same terms in a second closing subject to stockholder approval.

So that’s 97.5MUSD in 2005. I don’t see anything else prior to the acquisition. The final data (used for the table shown above is then):

1999 960,000USD (600,000GBP*1.6)

2001 19,040,000USD (13,600,000GBP*1.4)

2004 14,400,000MUSD (8,000,000GBP*1.8)

2005 97,000,000USD

2007 acquired for 600,000,000USD

It would be nice to get some idea of the valuation at various points. UK companies also publish share ownership. So we can roughly work this out. In the first filing in 1999 Abingworth owned exactly 50% of the company. For which they seem to have paid ~1MUSD. Valuing the company at 2MUSD. They received an additional 1.5MGBP from Abingworth sometime in 2000 increasing the stake to 60%. Valuing the company at up to 30MUSD (depending on how exactly these two investments were structured).

After this round a bunch of people jumped in. Amadeus, Oxford Bioscience, mRNA Fund II, Scroder Ventures, SITCO, and SV all invested in a 20.04MUSD round. The 2002 return shows that Abingworths stake dropped to 47%. I don’t think Abingworth put much in, so I’d guess the valuation here was around 50MUSD. After this round there was a round in 2004 which Amadeus led. If I’ve worked the numbers correctly then before/after that round ownership broke down as follows:

It looks to me like Amadeus put in basically all the cash. Other parties were largely diluted, though put in a small amount to help maintain stake. So of the 14.4MUSD I’m going to guess Amadeus put in 12MUSD. I should be able to work it more precisely but that ballpark, with a ballpark valuation for the company in the 30 to 60 million mark.

Keith over at OmicsOmics has an interesting write up on the recently announced Illumina Firefly system. As always, his views are interesting and informative, however he makes one assumption that I think may not be entirely warranted:

“The name is fun, though a bit curious given that the goal is to not use any light”

Here’s what I know about Firefly, which all comes from GenomeWeb via OmicsOmics:

• Uses a one-channel version of SBS chemistry.
• Based off CMOS sensor technology that Illumina acquired when it bought Avantome.
• Commercialize in second half of 2017.
• Costs <$30,000, cost per sample ~$100
• Library prep module uses digital fluidics, can process 8 samples in parallel.
• 1.2 gigabase per run.
• Uses nanowells that enable patterned clusters of DNA deposition.

From this I infer the following:

• It’s not a single molecule approach (note clusters).
• The chemistry is similar to their current chemistry.
• Uses a sensor that detects a single intensity type (i.e. no active filters, or differing illumination).
• Sequencing happens on a semiconductor somehow.
• May or may not have something to do with Avantome.

The pricing, cost per sample, and release data doesn’t express much about the basic technology. They are purely based on market factors. As is the current Illumina pricing.

Chip structure from the IEEE paper.

My initial guess based on the above data was that it’s an optical approach, using the existing chemistry on an integrated silicon “lab-on-a-chip” style sequencer. So I had a look around, first thing to do is a patent search. I couldn’t find anything directly relevant. But a bunch of their patents contain the following boilerplate statement:

In some embodiments described herein, detection of the signal, such as light emitted form conversion of ATP and luciferin, or light emitted form a fluorescent label, is detected using a charge coupled device (CCD) camera. In other embodiments, a CMOS detector is used. Detection can occur on a CMOS array as described, for example, in Agah et al., “A High-Resolution Low-Power Oversampling ADC with Extended-Range for Bio-Sensor Arrays”, IEEE Symposium 244-245 (2007) and Eltoukhy et al., “A 0.18mm CMOS bioluminescence detection lab-on-chip”, IEEE Journal of Solid-State Circuits 41: 651-662 (2006).

TL;DR: We can use a CCD camera (like current instruments) or we might use this thing talked about in the paper they reference. Here’s part of the abstract:

The paper describes a bioluminescence detection lab-on-chip consisting of a fiber-optic faceplate with immobilized luminescent reporters/probes that is directly coupled to an optical detection and processing CMOS system-on-chip (SoC) fabricated in a 0.18um process. The lab-on-chip is customized for such applications as determining gene expression using reporter gene assays, determining intracellular ATP, and sequencing DNA…

There also a nice presentation with the same title here. So what happened to the authors of the paper? Well the paper thanks Mostafa Ronaghi (current Illumina CTO via Avantome and NextBio). And its first author is Helmy Eltoukhy. His LinkedIn profile says he went on to become President and CEO of Avantome, before joining Illumina working on “Semiconductor-based Genetic Analysis Systems”.

Of itself I think this is enough to suggest that they’re looking at a photonics based lab-on-a-chip system using labels likely based around Bioluminescence.

I also came across this old job advert for a “Staff Development Engineer – Silicon Nanophotonics at Illumina” requirements include:

• Deep understanding of near field optics, physical principle of micrometer and nanometer scale silicon photonic devices
• Solid understanding of CMOS devices and semiconductor fabrication processes
• Experience with MEMS, biomedical equipment & devices, lab-on-chip

Finally, as I briefly mentioned above, those chips don’t use fluorescence like current systems. They use bioluminescence… you know… like fireflies… Hmmm? Get it?

Now onto some more speculation…

The analog and mixed signal circuitry, seems very similar to what you’d find in the photo-diode array industry. In a CCD, there’s typically a single amplifier and ADC, which is used to read out all the pixels. Photodiode array chips however often have an amplifier and ADC per sensor. You can typically find 128 channel integrated acquisition chips like the DDC1128. They operate at a few KHz bandwidth though and are used in things like CT scanners. The paper suggests they have a higher bandwidth. I’m not sure why this is necessary (see the table), but is seems kind of fun!

They use a patterned flowcell like their current instruments. Patterned flowcells no doubt simplified primary data analysis massively. In my experience, image analysis was always the most costly part of the primary data analysis process. Patterned flowcells should have made it much easier to extract intensities from clusters. The sensor per cluster approach simplifies data analysis by another order of magnitude, throwing out image analysis completely and allowing basecalling to from intensities directly.

And the basecalling process itself should be much simpler. There’s no crosstalk between lasers to compensate for, unless there are new signal artifacts your really just left with phasing to correct for. How much compute you require will depend on how many sensors you have. Let’s try and run the numbers.

We know the output is 1.2Gbases per run. Illumina read length tops out at 2x300bp. 1.2Gb/300/2=2 million. So if the above makes sense, we’re looking at a 2 million sensor chip. I have no reason to believe that isn’t feasible. From the IEEE paper, it looks like the pixels and acquisition electronics take up about equal space, so in terms of real-estate this is similar to a 4megapixel CCD. They might however be doing the acquisition differently. Making the die smaller than this.

So the chip needs to be as big as the flowcell. It would be interesting to know how big the die needs to be to accommodate 2 million nanowells and if this a sensible number.

You can probably work out the nanowell size from the number of reads per lane and size lane size. The illumina spec sheet contains the figure to the right which shows a SEM image,  but no units.

You have to be careful with pdfs though, as they can contain vector graphic as well as bitmap data. And vector graphics… over bitmap data. Google indexes all the images in the pdf even if they’re not visible.

Turns out that clipped SEM image is just masked in the pdf. Here’s the full SEM image which helpfully shows us the scale:

So a reasonable guesstimate is that the current wells are 500nm in diameter. Some futzing around on packomania tells me I can pack about 110 500nm circles into a 5000nm square. So I’ll need about 18181 of those squares, or 670 microns. A tiny about of space. I’d guess they’ll need more space than that, but the message is I can’t see the well size being a limiting factor.

Lets run the numbers again using the pixel size from the paper (240microns). If that were the case you’d need something like a 339mm squared die. Way too big. Those are pretty huge pixels though. For a reasonable sized die they must be using something much smaller. A pixel size of 10microns or smaller would be feasible I think, these are common in CCD cameras. The smaller the die, the cheaper. It’s unclear if the die is a consumable or not, but it’s certainly a possibility.

A 2 million sensor chip therefore looks like a reasonable guess. For my prior experience writing Illumina basecallers, basecalling 2 million intensities should be quite feasible on a low power processor, similar to what you’d find in a mobile phone.

So this is what I suspect we’re looking at, a small bioluminescence semiconductor sequencing platform which could potentially be very cheap. I may ponder a little more on what a single channel chemistry looks like, and the overall system architecture.

If the above is correct it’s a solid, iterative move development which is characteristic of Illumina. Illumina’s execution has always been excellent (and by excellent I mean better than everyone else). They’ve also always incrementally improved on their existing platforms, introducing chemistry improvements, new flowcells, and better acquisiton systems, slowly. This is also an incremental improvement in many ways. It retains all of their developments in chemistry but introduces a new labeling scheme and on die readout.

It is likely to enable them to produce a drastically cheaper sequencing platform, if they choose to go that route. However I don’t think it will enable massively longer read lengths. It may also prove difficult to scale the platform much beyond a few Gigabases and they are currently limited to a single read per well.

Disclaimer: I have to finish up with a disclaimer. I have stock in sequencing companies. I’ve worked for sequencing companies. This post is based on the information I’ve been able to dig up publicly (and is all I know that’s relevant anyway). This is all speculation, though I’ve tried to assess the technology on its merits.

I need a replacement for the battery cover on my calipers. It appears to be slightly different from those available on the Mitutoyo site (and the shipping cost is crazy). I modeled it and sent it out to fab in SLA. Will see how it turns out!

mitsu_bat

A collection of PCBs from these week. Also see Quadcopter 3D files and r2 of the PunkSeq10.

A esp1 to USB interface board:

An experiment in odd shaped PCBs (solar lantern):

Another odd PCB:

And another one, for experimentation: