Changing Pharma's DNA: An Interview with Paul McKenzie of BMS

Sept. 11, 2007
The vice president and general manager explains how BMS is harnessing the power of simulation, process analytical technologies and high-throughput technologies to make the plant operator's job easier, while reducing release and cycle times

Several years ago, former FDA Commissioner Mark McClellan challenged pharma to catch up with potato chip and soap flake manufacturers by modernizing operations and applying technology more effectively.

As risk-averse drug companies wrestle with the issues presented by manufacturing science, Bristol-Myers Squibb (BMS) has spent the past nine years quietly building a platform based on open control (particularly ISA’s S88 standard). The new platform utilizes automation and IT in ways that are novel to pharma, harnessing the power of simulation, process analytical technologies (PAT) and high-throughput technologies to improve process understanding, empower users and reduce cycle time.

BMS Vice President and General Manager Paul McKenzie’s goal is to have a single language for automation and IT that will allow data to flow much more efficiently between different operations, from the R&D lab, to QA and QC, to the plant floor.

S88 is often a hard sell in pharma, although many manufacturing engineers who have applied the standard appreciate its logic and agree that it simplifies operations, commissioning and validation. It requires stripping definitions of batch operations to their most basic level so that, for instance, the instruction “charge” will apply to a biotech, small molecule, or fill and finish plant. This is no easy feat, but believers say the front-end efforts pay off.

BMS has focused on applying S88, not in manufacturing, but to other areas further up the chain, improving the efficiency of drug development by getting scientists in R&D to understand the value of open control. “Rather than go to the fifth decimal place on the plant floor, I’d rather go to the first decimal point in the lab,” McKenzie told attendees at the World Batch Forum last May. Modeling is an extremely important part of the plan, and using models to predict, a priori, what will work in manufacturing, rather than learning the hard (and expensive) way, through wasteful “practicing.”

McKenzie and his team view S88 as much more than a mere standard. It’s really a DNA for the 21st century drug industry, McKenzie explains. “It provides a common vocabulary that can be used from the drug development teams all the way out to the commercialization teams,” he says. “It can be used as a common philosophy, not only in what historically would have been considered manufacturing events, but also in the lab events that sponsor manufacturing.”

BMS began working with S88 in small-molecule API development, as a way to improve tech transfer between R&D facilities in New Jersey and API manufacturing pilot plants in Ireland. Taking this approach for these construction projects removed roughly 9,000 signatures from the design and validation documentation. Now, the platform is being piloted in Puerto Rico for a fill/finish line, and rolled out in BMS’ new biopharmaceutical facility in Devens, Mass.

The key to success, says McKenzie, is not to think of automating for its own sake. “As Lynn Craig of the World Batch Forum has put it, S88 makes you think about where you should spend your automation money. It doesn’t mean everything should be automated,” he explains. “You must ask yourself: Where will a control valve or mass flow meter make the most sense? Where is the highest level of redundancy or unreliable operations? The ultimate goal is reducing data and enhancing its quality.”

Similarly, a paperless plant should not be the goal. “Electronic batch recordkeeping (EBR) is the natural outcome of a good execution plan, but it isn’t why you actually automate,” McKenzie points out. “We wanted to make the plant operator’s job easier and reduce release and cycle times.” Drug companies may be approaching EBR the wrong way, McKenzie suggests. “A lot of companies are migrating, in small steps, from the paper world to a paper-on-glass world, instead of sitting back and asking ‘How do I want to fundamentally change how I operate on the plant floor?’ ” he says. “S88 recipe execution allows you to address these questions. Good things come out of that, like EBR and reducing cycle times.”

In this interview, McKenzie discusses his vision, the origins of the platform and his views on pharma’s future.

PhM – Many of our readers are unfamiliar with S88, or consider it too cumbersome to use effectively. What led you and your colleagues to the standard?

PM – In 1998, we had the opportunity to build a new research pilot plant. I went out to benchmark, frankly, against everyone but pharma because pharma doesn’t use process control all that well.

PhM – Why is that?

PM – Historically, pharmaceutical innovation has been perceived to be at the chemistry and biology level. Engineering skill sets and applications and process control expertise have been out of its mainstream, at least compared with the petroleum and chemical industries.

As a result, pharma may do very well at introducing new medicines to the market, but it has done so relatively inefficiently.

PhM – Is automation based on S88 a prerequisite for efficient operations?

PM – At BMS, our vision has been “lab to plant,” where you work from the lab, collect the fundamental data you need, but from day one you’re using the vocabulary that is needed for targeting all lab development to successful commercial manufacturing implementation. This means not only higher quality drugs, but more efficient and environmentally safe processing.

PhM – How are you moving toward that vision?

PM – In 1998, we decided to use S88 to build a new small molecule pilot plant and a commercial API manufacturing plant. We harmonized across those two projects the S88 vocabulary, S88 approach, equipment P&IDs and codes. That gave us significant savings. Since those projects, we’ve used the platform when we renovated additional pilot plants in New Brunswick [N.J.], and we’ve moved it into our earliest stage kilo-scale lab. So, from our initial scale-up, we were collecting data in an S88 mindset.
More recently, we’ve launched a Manufacturing Operating Philosophy, which establishes a standard architecture that we’re starting to install in API and small molecule plants, in our bio plants, and in finished pharma product plants on both the development and commercialization side.

PhM – So the efforts were initially R&D-focused?

PM – We’re probably the first recipe-driven R&D organization where all pilot plants, whether for API or drug product, are based on S88. Most companies work on implementing the concepts in manufacturing first. Typically, R&D says, “That sounds like too much automation for me, and we will lose creativity and flexibility.” The reality is that if you build recipes right, you don’t lose that flexibility.

Paul McKenzie and his team at BMS are applying S88 early in the process, basing all their pilot plants on the standard.

PhM – What is left to accomplish?

PM – We have a lot of work to do before we fully achieve the vision of S88 at our commercial plants. We also want to bring the concept into the analytical labs to bridge the gap between them and the plant and reduce overall cycle time — a combination of plant floor exercise, release exercise, and all the analytical testing that goes with them. The more these efforts can be orchestrated together, the better off we’ll be . . . [and] the faster we’ll get more life-saving medicines to the patients who need them.

PhM – How did you ensure buy-in from top management?

PM – For the initial pilot plant and the commercial plant, the buy-in came from the fact that we were investing significant amounts of money — $180 million for the pilot plant and $525 million for the commercial plant — so it was important to make them run more efficiently. We took data from other industries to benchmark, and then did a small prototype. We evaluated results at that small plant before committing any automation dollars to larger plants. We also made sure that engineers came in from overseas to visit New Jersey operations, and that U.S. operators visited overseas plants so that everyone understood each other. Corporate buy-in came from the success of our small-volume prototype.

Vendors offering relevant solutions include:

PhM – Was there resistance on the plant floor?

PM – There was a challenge. In this industry, we process control engineers need to better articulate the difference between establishing the infrastructure required for automation, selecting an automation solution, and extending that solution. Often all three efforts wind up being tied to one dollar figure.

PhM – How are you applying high-throughput technologies outside of the drug discovery arena?

PM – The industry made a fundamental shift in the 1990s, when it adopted high-throughput screening for drug discovery. We asked how we could utilize the same technologies for R&D, to better develop the chemistry once we had a target molecule. Now, instead of doing things in 300-ml round-bottom flasks, you do them in 96 wells. We have had tremendous success using high-throughput techniques to optimize lab bench chemistries, so that we run 96 experiments at once and do multiple parametric studies of such factors as temperature, pressure and concentration. We collect fundamental process data at a speed that would be unimaginable using the old methods.

PhM – Will you be applying the same concepts to other areas?

PM – Today, the analytical testing processes required for release of a manufactured product tend to be very manual. People make the solutions by hand, pipette small volumes, and follow a very intricate methodology instead of automating. It isn’t that the industry hasn’t tried bits and pieces of automation, but overall, pharma companies usually reach the consensus that “we need more people in our analytical labs; these methods need to be executed by people.” I’d like to challenge that assumption. There’s a lot of opportunity for automation in this analytical lab bench space that we haven’t capitalized on yet, simply by taking what drug discovery did and putting it in the lab.

PhM – How are you doing this at BMS?

PM – Within some facilities, we have put the equivalent of an MES (manufacturing execution system) in the lab, providing analysts with electronic notebooks to allow them to access all the data and move through it. The next natural extension will be placement of a “batch engine” to run analyses and coordinate data through liquid handlers and integrated analytical and PAT instruments.

PhM – You mention using an “MES equivalent.” Are there other parallels between requirements for the plant floor and the lab?

PM – I see a lot of similarities between the plant floor’s batch engine, which tends to be DCSs and PLCs, and its manufacturing engine (MES), and what could exist in the lab. I hope that we will eventually have something like a batch engine within the lab that would be a high-throughput network, a distributed analytical network (similar to a DCS). The lab or manufacturing engine would function on top of that layer, scheduling all samples and removing people from processes where it makes sense to do so.

PhM – Is there a danger in automating too many lab processes?

PM – There is still an absolute need for people, but do you really benefit from having someone pipetting the same sample 300 times a year, when the job could be done more accurately and efficiently by a robot and computer? There are huge opportunities for utilizing the same S88-structured vocabulary within the analytical space that would improve quality departments’ connection with the plant and other operations, and help reduce overall cycle time.

The goal at BMS is to better use the data collected. They have concentrated on modeling data to move from the lab bench to pilot plants.

PhM – What is PAT’s role in your plans?

PM – On the small molecule side of the business, we invested significantly in PAT devices to help us understand what’s going on the plant floor without having to track everything via individual lab samples. Two of the main analytical techniques we’re using for PAT are Raman and NIR. We’re using Raman to release raw materials for identity testing so that a warehouse operator can identify materials right on the warehouse floor. We’ve also used Raman for crystal polymorph transformations, in cases where you have a very sensitive transformation from crystal type A to crystal type B in a 1,000-gallon reactor.

Using Raman, testing can be done in situ in the reactor in a matter of seconds, providing a complete profile of a transformation in close to real time. In these situations, you can’t take the sample and run to the lab and wait for the results.

PhM – Are you developing new PAT methods or evaluating new sensors?

PM – We’ve worked with Sandia National Laboratories on small gas chromatography-on-a-chip instrumentation, which we would use in some of our reactors. CPAC (the Center for Process Analytical Chemistry at the University of Washington) is doing some great work in this area.

PhM – How routinely is PAT being used, and are you piloting the high-throughput methods in the analytical laboratories yet?

PM – We use NIR and Raman routinely in pilot plants, as well as Lasentec for crystal sizing, and have transferred technology to the commercial area over the past few years. We’re currently piloting the high-throughput analytical concept in our stability labs, for dissolution and solubility and physical testing of solid dosage forms and API. Now it’s just a matter of a culture shift going forward. Can we get to productivity gains that discovery saw in the early 1990s by introducing these techniques to the drug development and commercial arena?

PhM – What is the major challenge with applying PAT in pharma?

PM – Clearly, it’s data management and integration. As an industry, we need to be able to distribute analytical equipment in our labs just as we do the temperature and pressure probes and other instruments in our plants. If this distribution were optimized, the industry could change the way it operates. Whether an analytical instrument ends up on a lab bench, next to a bioreactor or inside a bioreactor shouldn’t matter if we build the right data highway.

PhM – Does pharma need to bridge its lab and plant floor cultures?

PM – We need to move the lab space to the next level, and that’s where we have an industry divide. Historically, there’s not a lot of cross-fertilization unless it’s really forced, but there could be many synergies to explore. Other industries have already capitalized on them. I worked at DuPont early in my career, and technicians at its facilities not only operated the plant but also ran lab samples. Today, such hybrid technicians are still unusual for pharma.

PhM – We hear that many QC departments are still reluctant to try PAT or other methods of doing things. What will be needed to change that mindset?

PM – Most of culture shift needs to come from R&D. QC labs are charged with maintaining product quality and they don’t have time to tinker with new technologies. If R&D embraces these methodologies, QC labs will accept them. Quality labs are already up to their eyeballs in work. We can’t expect QC lab directors to go out and invest in PAT. PAT needs to come with the new processes as they are filed.

PhM – How is BMS applying simulation today?

PM – We’ve started using simulation in small-molecule development. Here, we’ve concentrated on how to model data, to move from the lab bench and the 96 wells to glass plant with kilos to pilot plant at hundreds of kilos to commercial scale with thousands of kilos. Our goal is to better use the data we’ve collected. As a result, we’ve partnered with vendors so that our data historians, which are structured in S88, can bring data and feed fundamental models for distillation, exothermic reactions and separations. This allows us to predict, once models are built, what a process would look like at Scale A, B or C.

PhM – Can you give any examples of where you’ve used simulation?

PM – We recently transferred a product from New Brunswick R&D into manufacturing. It was a very exothermic reaction, and we wanted to optimize parameters so that we had just the right temperature and concentration profiles. We used models to predict the best agitator and mixer regime for the commercial plant, and modified the agitator based on those models. We are now two years into manufacturing that product, without a single batch failure. Similar mixing studies are being done on the biotech side to optimize shear and other forces. This work is in its early stages. Several vendors are getting involved in biotech-oriented modeling, but vendors can only mature as industries adopt their technologies. They don’t want to build in functionality until the industry is at the point where it will adopt it — it’s an evolutionary process.

PhM – Are pharma companies focusing too much on the automation solution?

PM – Many plants pick a solution and extend in one small area. As a result, it gets very expensive and complicated in that one area. This effort (S88) should take place rationally, after the entire infrastructure is in place. Generally in pharma, we’re very good at taking small parts of plants to 105%, but leaving the rest of those facilities operating at 15%. Taking a slower, steadier approach and layering improvements is better than implementing them all at once.

PhM – Given the differences between small molecule development and biopharma, will you need to tweak the coding for each setting?

PM – Not really; some people could even use S88 API coding in biotech. People may customize too much for a given application. Ideally, you want to use the same modular piece of code more flexibly over many installations, regardless of what they make.

PhM – How do you ensure maximum flexibility?

PM – We’ve looked at the overall matrix of operations to ask how we could reuse the same code in multiple places. We’ve done some recent benchmarking and were very impressed with how some companies have been able to use multiple parts of code across API and biologics with very little change.

PhM – Going back to Dr. McClellan’s criticism, isn’t it time for industry to embrace change?

PM – The drug industry is embracing change, but we’re trying to figure out which path will be most productive. It’s refreshing that FDA is now encouraging change. More people in the industry now realize that we need to continue to ensure the highest quality, but increase overall productivity in manufacturing. As molecules move forward, we want to decrease cycle times, but to do that, we will need to rethink how we work.

PhM – Fears of revalidation appear to be stopping some modernization projects in pharma, but doesn’t S88 simplify the overall validation process?

PM – E55 (the ASTM standard for using PAT in pharmaceutical manufacturing) and S88 are the perfect marriage. With both in place, we could fundamentally change how we do IQ and OQ in the industry. For example, if you look at the history of validation, it has always been done on individual pieces of equipment. With a strong S88 model, you could move from doing IQ and OQ on individual pieces of equipment to doing it for an entire process recipe.

PhM – Are FDA’s inspectors up to speed with what Agency leaders have been advocating? Do they “get” S88?

PM – There’s a learning curve involved for everyone, but I’ve given some S88 training courses to new inspectors, and in many ways, S88-driven execution is easier for people to understand. If you can show good engineering, it should be easier to approve than a number of separate manual operations. It even opens up the possibility one day for inspections to be handled remotely.

PhM – Are there any precedents in other industries for what needs to happen in pharma today?

PM – I don’t touch my 401K assets, but I can model data to optimize retirement plans, etc. The same holds for plant data; we don’t want to touch it, but we need to use it for modeling, etc. Other industries are doing this — gas facilities and refineries are run remotely and this would definitely be doable in our industry. It’s just a matter of time.

PhM – Is the role of the process engineer changing in this industry?

PM – We’ll see an increase in the number of engineers working in the pharmaceutical industry, but one of the things that I like about working in this industry is the fact that there’s no single group that trumps the need for others. Engineers have played less of a role than they will, moving forward.

PhM – Is training adequate for engineers in pharma today?

PM – We have to change training in academic and industrial settings but we don’t know how yet. Some European institutions, such as University College Dublin, offer very practical unit operations classes using real plant equipment. My training, in contrast, was very theoretical — for example, measuring temperature profiles in a rod. There needs to be a fundamental shift in how we expose students to the global nature of business and technical operations.

PhM – How much of a reduction in cycle time have you seen so far at BMS, as a result of all these efforts?

PM – We’re seeing release time in R&D move from a weeks-to-months timeframe to within days. We’re looking hard at overall cycle times, from the time that raw materials are purchased, onward. Over the next two to three years, as the new bio plant comes on stream, we expect to show dramatic changes in cycle times.

No Paper on Glass at Devens

A showcase for BMS’ automation strategy will be its $750 million biologics manufacturing plant in Devens, Mass. that will employ roughly 400 people when operational in 2011. Its first product will be Orencia, which is used to treat rheumatoid arthritis.

This facility will have 120,000 L of bioreactor capacity capable of concurrent multi-product production. The plant will leverage the BMS Manufacturing Systems’ architecture to integrate the manufacturing execution, process control, building automation and lab systems with the ERP (Enterprise Resource Planning) platform. Thus, one data center at the facility will host manufacturing, lab, process control, building automation, security and business systems.

The first phase of construction involves four main buildings, including the manufacturing structure that will house six 20,000-L cell culture vessels and one purification train, explained senior principal engineer Tony Fenn, who summarized the efforts at the Operational Excellence conference in Philadelphia on August 3.

The facilities are due to be operationally complete in 2009, with submission for regulatory approval timed for 2010 and operations at the site beginning in 2011.

BMS engineers were determined not to install a manufacturing execution system on top of manual/paper-based processes. “Paper on glass was not an option,” says Fenn. Instead, the goal was to emphasize plant floor performance and standards to make operators’ jobs easier, allowing them to focus more on the batch process and to refer to a single system.

Reporting is done by exception, Fenn explained, and overall the goal is to minimize human intervention and enable integration across manufacturing, quality and plants.

Key features will be:

  • Electronic batch recordkeeping
  • Quality review by exception that will expedite batch review and release
  • Electronic Work Instructions to reduce errors
  • Deviation and change control systems
  • Technologies such as IP telephony and wireless controllers.

Construction is progressing well, McKenzie says, and the last piece of steel went up on August 24. Plant building will be typical ‘bricks and sticks’ constructed in the field, but assembly of reactors, buffer and media hold and charge tanks will be done on superskids.

About the Author

Agnes Shanley | Editor in Chief