ICH M4Q(R2) And M16 Modernize The CTD For Digital Submissions

A conversation between Katie Duncan at GSK and Life Science Connect's Jon O'Connell

ICH M4Q(R1) — the quality module of the common technical document — was written for a pre-digital era of regulatory CMC review.

Today's submissions are vastly more complicated, and they're expected to support much faster regulatory assessments. To catch up with regulatory expectations and greater data demands, the International Council for Harmonisation (ICH) set out to overhaul M4Q(R1) and develop the M16 Structured Product Quality Submission, a counterpart standard, to begin closing the gap.

To help us understand the changes to M4Q and how M16 enables structured submissions, Katie Duncan, a CMC policy and advocacy director at GSK and former FDA quality assessor, offered to answer some of our questions. She's scheduled to dig into the latest updates next month at the 2026 ISPE Annual Meeting & Expo and gave us a preview of her talk.

Beyond the fact that it's simply 24 years old, what about ICH M4Q(R1) needed updating?

Duncan: M4Q(R1) was not built to handle the dramatic technological or scientific advancements of the past two decades. First, M4Q(R1) was written in a far more analog time, one still dominated by PDFs and paper-based submissions. Second, submissions have gotten significantly more complex in recent years, as development of modalities like antibody-drug conjugates and combination products like autoinjectors have taken off.

In addition, M4Q(R1) was developed before many modern ICH quality guidelines, including ICH Q8-Q14. The linear narrative structure of M4Q(R1) makes sense when reviewers would read printed documents from front to back, but it doesn’t stand up to how data are generated, managed, and reused today. M4Q(R1) needed modernizing to enable greater knowledge sharing and management and to address the volume and complexity of today’s regulatory submissions.

How do ICH M4Q(R2) and the emerging M16/SPQS relate to each other?

Duncan: M4Q(R2) and M16 are both necessary to the modernization of regulatory submissions. I think of M4Q(R2) as defining the architecture of the submission (how the content is organized), while M16 provides the underlying data standard, defining how that content gets packaged and exchanged electronically as a Structured Product Quality Submission. In other words, M4Q(R2) is the framework and structural blueprint and M16/SPQS is the technical enabler, the transport, and submission mechanism built to carry it. In order to realize the ultimate goal of a fully digital regulatory ecosystem, we need a clear, organized framework for the structured, standardized data.

Earlier this year, BioPhorum and other groups warned that industry lacked time and instructions to get from M4Q(R1) to M4Q(R2). Have we cleared that hurdle? If not, what steps are we still struggling with?

Duncan: I think we’ve made tremendous strides in understanding the scope of the transition from M4Q(R1) to M4Q(R2), but hurdles still remain. Most companies, GSK included, are trying to figure out what this transition looks like in practice. We are trying to figure out how to leverage existing templates and systems and what additional templates and systems are required.

Additional hurdles include trying to understand the timeline for implementation and preparing for the potential of maintaining two dossier structures if regions adopt M4Q(R2) at different times. We’re also trying to navigate how legacy content will be managed. I expect many of these issues to be clarified as work on M4Q(R2) progresses and training materials, mapping documents, and examples are further developed and refined.

From your time with FDA as a quality assessor, what made CTD-format submissions slow or difficult to review? How does machine-readable data solve those challenges?

Duncan: When I was at the FDA, a large part of the overall review time was spent trying to locate data and then recreate that data in our reviews. I did a lot of “copy-paste” from a PDF in a dossier to my Word document assessment. For myself and my fellow assessors, many information requests were the result of not being able to find information in the dossier or trying to understand inconsistencies between different sections of the dossier.

When everything is free text, assessors must read, interpret, and manually verify consistency every time. Machine-readable data helps both industry and regulators to do all these tasks more rapidly. Machine-readable data makes it straightforward to find data and information, freeing the reviewer to spend time focusing on interpreting and evaluating that information.

What do AI-supported CMC submissions look like in practice? How do AI and humans divvy up the work?

Duncan: I think we’re just starting to figure out what AI-supported CMC submissions look like and are navigating the advantages and disadvantages of this powerful tool. Today, AI serves a mostly assistive role: helping to assemble and cross-check content, writing first drafts based on underlying data, and identifying inconsistencies between sections.

Because it excels at pattern recognition and cross-referencing, AI can dramatically reduce the time spent authoring and quality control testing. On the other hand, humans create the scientific and regulatory strategy, interpret and contextualize the data, and revise the AI-generated first drafts. Importantly, humans bear the ultimate accountability for the final submission. I don’t see this division of labor dramatically changing as the tools advance; rather, the questions center around how much oversight the AI requires.

A single harmonized quality module covering small molecules to ATMPs runs counter to a more recent push toward modality-specific guidance. Must companies now build in modality-specific workarounds?

Duncan: I think of it less as modality-specific workarounds and more as modality-specific adaptations of a common structure. M4Q(R2) provides a generalized quality model that provides a framework that is as applicable to small molecules as it is to newer modalities like ATMPs and oligonucleotides. The question becomes which elements of the framework are relevant to the modality and how they are described given the modality’s specific manufacturing, quality, and control considerations.

The M4Q(R2) framework allows for greater freedom to describe the modality-specific, holistic control strategy. Industry and regulators will need to keep pace with that flexibility, generating templates and trainings that are appropriate for the new structural paradigm.

Do you envision a future where AI-written dossiers are the norm and expected by regulators? Or will they be a nice-to-have for those who have the time and attention to build out those systems?

Duncan: Yes, I think the question is more of “when” and not “if.” From all the interest and investment we’re seeing among pharmaceutical companies, the industry is moving to a place where AI-written dossiers are the norm. The timeline for broad adoption across the entire range of regulators and pharmaceutical companies is far less clear.

I think the transition to AI-written content will be more gradual than today’s hype would imply. As AI-assisted authoring becomes increasingly fast and more reliable, more and more companies will turn to these systems and their prevalence will increase. I predict it will be analogous to the way electronic submissions replaced paper submissions: we will encounter a tipping point where the efficiency gains for both regulators and industry made the novel more desirable than the traditional.

About The Expert:

Katie Duncan is a CMC policy and advocacy director at GSK. Previously, she served as a pharmaceutical quality assessor and, before that, a staff chemist in FDA's Office of Pharmaceutical Quality. She received her Ph.D. in organic chemistry from the Scripps Research Institute.