BRENNEN

Industries / Health / Pharmaceuticals / Contract drug manufacturing

Moving a process between plants

This is technology transfer: handing a working process, with its documents and know-how, from a drug company or one plant to another, and getting regulators to accept the change.

5

Milestones achieved, from 2011 to 2021.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2011

    Rules for handing over

    The World Health Organization's (WHO) Expert Committee on Specifications for Pharmaceutical Preparations published guidelines on transfer of technology in pharmaceutical manufacturing (Technical Report Series 961, Annex 7). They name a sending unit and a receiving unit, and say a transfer means both the documents and the receiving unit's shown ability to run the critical steps. WHO noted that regulators had published little on the subject, and that guidance on transfers within a company came from the International Society for Pharmaceutical Engineering's (ISPE) 2003 Good Practice Guide. ISPE issued its third edition in December 2018, and WHO revised its own guidelines in 2022 (Technical Report Series 1044, Annex 4).

    extranet.who.intwho.intispe.org

    Credit

    WHO; ISPE

  2. 2019

    Changes planned in advance

    ICH's regulators adopted guideline Q12 on 20 November 2019. It lets a company name "established conditions", the details of a process that need approval to change, and file a post-approval change management protocol (PACMP): a plan, agreed with regulators beforehand, for a future change such as a new site. The EU had used such protocols since 2010. In March 2020 EMA said that Q12's product lifecycle management document, and some of its ways of setting established conditions, did not fit EU law, which takes precedence. The EU revised its rules on changes from 2025, with new guidelines applying from 15 January 2026.

    Lessona shared guideline works only as far as each region's law allows.

    ema.europa.euema.europa.euema.europa.eu

    Credit

    ICH; EMA; European Commission

  3. 2020

    A transfer in months

    In May 2020 Moderna hired Lonza to make the active ingredient of its COVID-19 vaccine, mRNA-1273, first at Portsmouth, New Hampshire, then at Visp, Switzerland. Lonza said its existing clean rooms would be ready for the transfer in June 2020 with no major construction, and that large-scale production began at Portsmouth at the end of September 2020. In the EU, a change adding Lonza's two Visp plants as makers of the vaccine's active substance won the backing of EMA's committee on 18 March 2021.

    Lessonusing finished, empty clean rooms took building work out of the transfer.

    bioprocessintl.comlonza.comema.europa.eu

    Credit

    Moderna; Lonza

  4. 2021

    One vaccine, two listings

    On 15 February 2021 WHO listed two versions of the AstraZeneca/Oxford COVID-19 vaccine for emergency use: one made by SK bioscience in the Republic of Korea and one by the Serum Institute of India. WHO assessed both in under four weeks, and its listing process includes inspecting the plants where needed.

    Lessonto a regulator, a product made at a new plant is a new product until that plant's output has been checked.

    who.int

    Credit

    AstraZeneca; University of Oxford; SK bioscience; Serum Institute of India; WHO

  5. 2021

    A hub for mRNA know-how

    WHO and the Medicines Patent Pool (MPP) announced an mRNA technology transfer programme on 21 June 2021, to build the ability of low- and middle-income countries to develop and make such vaccines. Its hub is in Cape Town: Afrigen Biologics, which developed its own COVID-19 mRNA vaccine technology, the vaccine maker Biovac and the South African Medical Research Council. The hub is to teach other plants the process, and on 1 May 2025 the programme had 14 manufacturing partners besides the hub.

    Lessonknow-how travels with the people who use it. A veteran interviewed for a 2024 review of the hub by Matthew Herder (Dalhousie University) and Ximena Benavides (Yale University) said a transfer "needs to be in the facility that's sending the technology out".

    who.inthealthpolicy-watch.news

    Credit

    WHO; MPP; Afrigen Biologics

Watch for

  1. Digital handovers

    BioPhorum, an industry group of biopharmaceutical makers, published "Digital Technology Transfer Playbook: Guiding Principles" in October 2026. It says structured data objects and digital product definitions can make transfers faster and more consistent than the static documents, spreadsheets and staff memory now used.

    Achieved whena drug company and a contract maker publish a transfer done by exchanging structured data, with its time and error count set against a document-based transfer.

    chemanager-online.com

    Claimed by

    BioPhorum, October 2026

  2. The mRNA hub's first transfers

    WHO says the hub completed its first transfer in September 2024, to Biovac, which is itself part of the hub, and that three more partners received the technology that year. Afrigen's chief executive, Petro Terblanche, said in June 2025 that the technology had gone to eight partners and that Afrigen was "on the cusp of receiving GMP accreditation".

    Achieved whenSouth Africa's medicines regulator lists a GMP licence for Afrigen's plant, or a regulator approves a product made with the hub's technology at a partner's plant.

    who.intwho.int

    Claimed by

    WHO, which co-leads the programme, and Afrigen Biologics, June 2025

Open

  1. 01

    A public site-change log

    A table of every manufacturing-site change approved for medicines authorised centrally in the EU, drawn from EMA's "procedural steps taken and scientific information after authorisation" documents. Each row gives the product, the variation number, its change codes (for example B.II.b.1, a new site for the finished product, or B.I.a.1, a new maker of the active substance), the opinion and decision dates, the sites named, and whether the change used an approved change protocol (PACMP).

    The brief
    Credit · unclaimed
  2. 02

    A transfer package from A-Mab

    A machine-readable technology transfer package for A-Mab, a made-up antibody whose full development story was published in 2009 as a 278-page case study. The package lists every process step, its parameters with ranges, which parameters are critical, and the quality attributes each one affects, in an openly published format. Then a test of how well AI models build the same package from the PDF.

    The brief
    Credit · unclaimed

Open 01 · the brief

A public site-change log

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A table of every manufacturing-site change approved for medicines authorised centrally in the EU, drawn from EMA's "procedural steps taken and scientific information after authorisation" documents. Each row gives the product, the variation number, its change codes (for example B.II.b.1, a new site for the finished product, or B.I.a.1, a new maker of the active substance), the opinion and decision dates, the sites named, and whether the change used an approved change protocol (PACMP).

Why it is still open

The figures in use are estimates. EFPIA, the European drug industry association, wrote in 2017 that adding a manufacturing site can take 3 to 5 years worldwide, against about 6 months at EMA. A 2023 trade column says change protocols cut approval times by around six months, without naming its data. The records exist: EMA's 75-page log for Moderna's Spikevax lists each site change, such as the group that added Lonza's Visp plants. I found no table that counts these changes across products.

efpia.eupharmaceuticalonline.comema.europa.eu

Built with what exists

EMA's public product pages and PDFs; the plant names gathered for "Who makes each antibody" under Biologics, which this log adds dates and change codes to; pdftotext and Python to parse the tables; a language model such as Claude to read site names where the page layout breaks, with a person checking; a CSV file in a public repository.

  1. First step

    Parse the Spikevax log and count its site-change variations by code and year.

  2. Done looks like

    All centrally authorised products covered, change codes parsed with at least 98% accuracy on a 200-row hand check, and published counts of site changes per year and per product type, with and without a change protocol.

Open 02 · the brief

A transfer package from A-Mab

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A machine-readable technology transfer package for A-Mab, a made-up antibody whose full development story was published in 2009 as a 278-page case study. The package lists every process step, its parameters with ranges, which parameters are critical, and the quality attributes each one affects, in an openly published format. Then a test of how well AI models build the same package from the PDF.

Why it is still open

Transfers still run on static documents, spreadsheets and staff memory, and BioPhorum's October 2026 playbook gives principles for digital transfer, not a public worked example. A-Mab, written by the CMC Biotech Working Group and placed in the public domain, is one of the few complete process descriptions anyone can use, and it includes a section on changing sites.

chemanager-online.commdsoar.org

Built with what exists

the A-Mab PDF (a CC0 copy is held in MDSOAR, Maryland's shared open-access repository); a JSON schema written for the purpose; Claude, Gemini and Mistral Large 4 (Note: "A trillion open parameters"); a spreadsheet answer key made by two people. The advance that makes this possible is language models that read long technical PDFs and fill a fixed table.

  1. First step

    Key in by hand the parameters of one step in A-Mab, such as the production bioreactor, and test one model on the same pages.

  2. Done looks like

    An answer key of every parameter and range in the case study, checked by two people, and published scores for at least three AI models on how many entries each gets right and how many it invents.