Environmental monitoring practices relevant to regenerative medicine manufacturing: Insights from a cross-sector survey of regulated facilities in Japan Authors: Mitsuru Mizuno, Tomoko Oono, Kana Takahashi, Kazuko Tsujimoto, Natsuki Tanigawa, Yoshinobu Nakamura, Hiroaki Nakamura, Kenji Fujii, Ichiro Sekiya Published: 2026 ABSTRACT AbstractIntroductionEnvironmental monitoring (EM) is essential for contamination control in regenerative medicine manufacturing, where aseptic handling and operator-dependent processes require adaptable yet robust strategies. Although EM is governed by shared regulatory requirements across manufacturing sectors in Japan, implementation varies considerably depending on product characteristics and manufacturing processes. However, quantitative investigations that position regenerative medicine within the broader landscape of regulated manufacturing sectors remain limited. This study aimed to characterize variation in EM practices across sectors in Japan and identify key factors driving these differences.MethodsA cross-sectional questionnaire survey was conducted among EM personnel at regulated pharmaceutical and regenerative medicine manufacturing facilities in Japan. Fifty-five respondents from six sectors completed a 55-item questionnaire. Fifty-four items were numerically encoded and analyzed using PCA, k-means clustering, PERMANOVA, LDA, and one-way ANOVA with effect size estimation (η2).ResultsEM practices differed significantly across sectors. Major differences included implementation of surface monitoring in Grade B/C-equivalent areas, frequency and approach of media performance testing, and timing of airborne particle monitoring. Regenerative medicine facilities showed a mixed profile, combining frequent media testing with variable monitoring coverage. Clustering further separated groups based on clarity of sampling rationale and microbial identification practices.ConclusionEM practices in regenerative medicine are partially aligned with standardized sterile manufacturing but remain heterogeneous. Identifying high-impact domains provides a basis for improving risk-based EM strategies, documentation, and training in regenerative medicine manufacturing.Keywords: Environmental monitoring, Pharmaceutical manufacturing, Cleanroom, Risk-based monitoring, Multivariate analysis, Cross-sectional survey Abbreviations EM GPT PCA LDA Introduction Environmental monitoring (EM) is a foundational element of contamination control in pharmaceutical and regenerative medicine manufacturing. It plays a crucial role in ensuring that cleanroom environments remain within microbial and particulate control limits necessary for aseptic drug and cell product manufacturing [[1], [2], [3], [4], [5], [6]]. In Japan, manufacturing facilities for these products are regulated under product-category-specific ministerial ordinances that share common underlying principles of quality assurance and contamination prevention. Although the regulatory designations differ by product type, the core requirements for environmental control are substantively aligned. Further, despite baseline expectations being defined by regulatory requirements, the practical implementation of EM programs varies widely across manufacturing sectors—from sterile small-molecule production to biologics, non-sterile drug products, and regenerative medicine facilities. These differences arise from divergent risk profiles, operational constraints, historical practices, and the degree of regulatory oversight applied to each sector [[7], [8], [9], [10]]. These differences are rooted in the distinct quality objectives associated with each product category. EM is critical to sterility assurance for sterile drug products. Further, regulatory guidance requires cleanroom classification with routine monitoring of airborne particles, airborne viable microorganisms, and surface microorganisms at defined locations and frequencies [4]. Non-sterile drug products and active pharmaceutical ingredients are subject to microbial limit specifications rather than sterility requirements, and the applicable regulatory expectations for environmental control are therefore less prescriptive in terms of cleanroom classification and monitoring scope [4,11]. Within this spectrum, regenerative medicine manufacturing represents a particularly distinctive context. Production processes often involve open or semi-open handling, intensive operator interaction, relatively small production scales, and starting materials that differ substantially from those used in conventional pharmaceutical manufacturing. These features can increase contamination risk and necessitate carefully tailored EM strategies. However, the diversity of manufacturing formats for regenerative medicine — ranging from open cleanroom operations to closed or semi-closed processing systems — disallows the possibility of a single EM approach that is universally applicable; hence, monitoring programs must be adapted to the specific contamination risks posed by each production workflow [[12], [13], [14]]. Despite the recognition of these challenges, systematic cross-sector comparisons that position EM practices in regenerative medicine within the broader landscape of regulated manufacturing remain limited. Previous studies have examined aspects of environmental contamination, monitoring technologies, and risk management in cleanrooms. However, most studies focus on single facilities, specific product modalities, or individual monitoring methods [[15], [16], [17]], thereby limiting the benchmarking of EM practice diversity across sectors. Moreover, knowledge regarding how facility type shapes operational decisions is also limited, including decisions regarding microbial sampling frequency, surface monitoring coverage, growth promotion testing (GPT) implementation, or rationale for particle-monitoring procedures. Furthermore, the increasing adoption of advanced therapies, particularly regenerative medicine products, has introduced new workflows and facility designs that may diverge substantially from traditional pharmaceutical EM models. Therefore, a comprehensive, sector-spanning assessment is crucial for clarifying whether EM practices exhibit shared patterns or if distinct operational profiles dominate within each industry type. This study addresses this gap by conducting a questionnaire survey of personnel engaged in EM activities at regulated manufacturing facilities across multiple product sectors in Japan. This study uses multivariate statistical approaches, including PCA, k-means clustering, PERMANOVA, and item-level analyses, to systematically evaluate how EM practices vary across facility types. It identifies the specific monitoring domains that contribute significantly to these differences. By placing regenerative medicine facilities within a broader cross-sector context, this study aimed to clarify which aspects of EM are relatively standardized and which remain heterogeneous. These findings may support risk-based optimization, documentation, training, and harmonization of EM programs in regenerative medicine manufacturing. Materials and methods Study design and participants The cross-sectional questionnaire survey targeted personnel involved in EM activities at regulated pharmaceutical and regenerative medicine manufacturing facilities in Japan. Respondents were recruited from facilities manufacturing sterile drug products (small-molecule and biologics), non-sterile drug products, active pharmaceutical ingredients (APIs), and regenerative medicine products. Participation was voluntary and anonymous. Additionally, no personal identifiers or institutional information were collected. Facility identifiers were not recorded, indicating that multiple respondents may have originated from the same site. Therefore, responses reflect individual-level practices and perceptions rather than site-level aggregated data. No restrictions were implemented to limit participation to one response per facility. A total of 55 individuals completed the questionnaire and were included in the analysis. Questionnaire structure The questionnaire comprised 55 items, covering airborne particle monitoring, airborne and surface microbial monitoring, media handling and testing, microbial identification, and equipment management (Table 1). We developed all questions and response formats to ensure clarity and applicability for personnel working in regulated manufacturing sectors. Table 1Questionnaire items and response options used in this study.The table lists all questionnaire items (Q01–Q55) with English item wording and corresponding response options; the English text represents translations of the original Japanese questionnaire.Table 1IDItem (EN)Options (EN)Q01What type of product is mainly manufactured at your site (or in the department you are responsible for)?(1) Sterile drug products (small-molecule) (2) Sterile drug products (biologics) (3) Non-sterile drug products (4) Drug substance (for sterile products) (5) Non-sterile APIs (6) biologics APIs (7) Regenerative medicine products (cell/tissue-based products)Q02Which department at your site is responsible for environmental monitoring, including planning and approval of the EM program?(1) QC (2) Manufacturing/Production (3) Engineering/Facilities (4) QA (5) OtherQ03Which department actually performs the environmental monitoring sampling work at your site?(1) QC (2) Manufacturing/Production (3) Engineering/Facilities (4) QA (5) Other (other departments, temporary staff) (6) External contractor (7) multiple selectionAirborne particlesQ04What particle sizes are measured by your particle counter?(1) 0.5 μm and 5.0 μm (2) 0.5 μm only (3) OtherQ05When are airborne particles measured?(1) Both at-rest and in-operation (2) In-operation only (3) At-rest only (4) Not clearly distinguished (5) OtherQ06What is the basis for determining the current EM sampling points for airborne particles?(1) Set per area in accordance with ISO 14644-1 (2) Set by location based on risk assessment (3) Combination of (1) and (2) (4) Unknown; defined in SOP (5) OtherQ07How is the delay time (stabilization time) for particle counting set?(1) Set for both at-rest and in-operation (2) Set for in-operation only (3) Set for at-rest only (4) Not set (5) OtherQ08What is the basis for the delay-time setting in both at-rest and in-operation conditions?(1) Set internally considering cleanliness recovery time (2) Set according to air change rate (3) Set, but the basis is unknown (4) Other; (5) No responseQ09What is the basis for the delay-time setting in in-operation conditions?(1) Set internally considering cleanliness recovery time (2) Set according to air change rate (3) Set, but the basis is unknown (4) No responseQ10What is the basis for the delay-time setting in at-rest conditions?(1) Set internally considering cleanliness recovery time (2) Set according to air change rate (3) Set, but the basis is unknown (4) Other (5) No responseQ11If no delay time is set, what is the reason?(1) Not enough time for sampling (2) No meaningful benefit (3) Not set, but the reason is unknown (4) Other (5) No responseQ12How often is the airborne particle EM monitoring plan reviewed?(1) Annually (2) Every few years (3) Not changed unless there is an issue (4) Other (5) UnknownQ13How is the sampling volume for particle counting set?(1) Calculated to meet the minimum sampling volume in ISO 14644-1 (2) Set to a uniform volume across all areas (e.g., 1 m3) (3) Defined in SOP (4) Other (5) UnknownQ14What is the suction flow rate of the particle counter used at your site?(1) 2.83 L/min (2) 28.3 L/min (3) 50 L/min (4) 100 L/min (5) Other (6) UnknownQ15Why was this suction flow rate selected for the particle counter?(1) Selected based on required sampling volume (2) Unknown; defined in SOP (3) Never considered (4) Other (5) UnknownQ16How is the sampling height for particle counting defined?(1) Defined in SOP (2) Not specified in SOP, but measured at a consistent height (3) Other (4) UnknownQ17What is the rationale for the selected sampling height?(1) Matched to process/working height (2) Set based on risk assessment (3) Defined, but the rationale is unknown (4) No responseAirborne microorganismsQ18Which airborne microbial sampling methods are used (active air sampling, settle plates)?(1) Both active air sampling and settle plates (2) Active air sampling only (3) Settle plates only (4) Other (5) UnknownQ19What type of active air sampler is used for airborne microorganisms?(1) Perforated (pinhole) type (2) Slit-to-agar type (3) Centrifugal type (4) Rapid microbial method (5) Other (6) Unknown (7) No responseQ20Which culture media are used for active air sampling?(1) Soybean–casein digest agar (SCD)/Tryptic soy agar (TSA) (2) Other (3) Unknown (4) No responseQ21When is active air sampling performed (at-rest/in-operation)?(1) Both at-rest and in-operation (2) In-operation only (3) At-rest only (4) Other (5) No responseQ22What is the basis for selecting sampling points for active air sampling?(1) Same locations as particle monitoring (2) Set by area based on risk assessment (3) Combination of (1) and (2) (4) Unknown; defined in SOP (5) OtherQ23Is a delay time set for active air samplers?(1) Set (2) Not setQ24If delay time is set, what is the basis for this setting?(1) Set internally considering cleanliness recovery time (2) Set according to air change rate (3) Set, but the basis is unknown (4) Other (5) No responseQ25If no delay time is set, what is the reason?(1) Not enough time for sampling (2) No meaningful benefit (3) Not set, but the reason is unknown (4) No responseQ26How often is the airborne microbial (active) monitoring plan reviewed?(1) Annually (2) Every few years (3) Not changed unless there is an issue (4) Other (5) No responseQ27How is the sampling height for active air samplers defined?(1) Defined in SOP (2) Not specified in SOP, but measured at a consistent height (3) No responseQ28What is the rationale for the selected sampling height for active air samplers?(1) Matched to process/working height (2) Set based on risk assessment (3) Defined, but the rationale is unknown (4) No responseQ29What is the exposure time per settle plate for passive air sampling?(1) Exactly 4 h (2) Between 2 and 4 h (3) ≤ 2 h (4) Varies depending on cleanliness grade or process (5) Other (6) No responseQ30What criteria are used to interpret results from settle plates?(1) Based on acceptance limits regardless of exposure time (2) Converted to a 4-h equivalent (3) In-house criteria (4) Other (5) UnknownQ31Which culture media are used for settle plates?(1) SCD/TSA (2) OtherQ32When is passive air sampling with settle plates performed (at-rest/in-operation)?(1) Both at-rest and in-operation (2) In-operation only (3) At-rest only (4) Other (5) No responseQ33What is the basis for selecting sampling points for settle plates?(1) Same locations as particle monitoring (2) Set by area based on risk assessment (3) Combination of (1) and (2), (4) Unknown (5) OtherQ34How is the sampling height for settle plates defined?(1) Defined in SOP (2) Not specified in SOP, but placed at a consistent heightQ35What is the rationale for the selected sampling height for settle plates?(1) Matched to process/working height (2) Set based on risk assessment (3) Defined, but the rationale is unknown (4) No responseSurface microorganismsQ36Which methods are used for surface microbial sampling?(1) Contact plates (2) Swabs (3) Rapid microbial method (4) Other (5) Unknown (6) Combination of (1) and (2)Q37How are sampling points for surface microorganism monitoring selected?(1) Based on risk assessment (2) Defined, but the rationale is unknown (3) Other (4) UnknownQ38Is surface microbial sampling performed in Grade A isolators?(1) Yes (2) No (3) No isolator available (4) Other (5) UnknownQ39… in Grade A biological safety cabinets?(1) Yes (2) No (3) Not available (4) Other (5) UnknownQ40… in Grade A RABS?(1) Yes (2) No (3) Not available (4) Other (5) UnknownQ41… in Grade B (operational ISO 7) areas?(1) Yes (2) No (3) No such area (Grade B/operational ISO 7) (4) Other (5) UnknownQ42… in Grade C (operational ISO 8) areas?(1) Yes (2) No (3) No such area (Grade C/operational ISO 8) (4) Other (5) UnknownQ43… in Grade D or ISO 9 areas?(1) Yes (2) No (3) No such area (Grade D/ISO 9) (4) OtherMedia handling & performance testingQ44Which department is responsible for incubation and reading of EM culture media?(1) QC (2) Manufacturing/Production (3) Engineering/Facilities (4) QA (5) Other (6) External contractorQ45Do you perform growth promotion tests (GPT) or equivalent media performance tests for EM media?(1) Performed in-house (2) Verified by certificate of analysis (CoA) at receipt (3) Unknown; (4) Outsourced to an external contractorQ46If performed, how frequently are media performance tests conducted?(1) Each batch/lot (2) Each purchase (3) Before each use (4) Periodically (5) Other (6) No responseQ47Which guideline is used as the basis for media performance testing?(1) Japanese Pharmacopoeia (JP) (2) United States Pharmacopoeia (USP)(3) European Pharmacopoeia (Ph. Eur.) (4) Unknown (5) OtherQ48What acceptance criteria are used for media performance testing?(1) Mean values of indicator organisms at adoption/approval (2) Use of quantified reference strains with predefined acceptance ranges (3) In-house criteria (4) Unknown (5) OtherMicrobial identificationQ49Do you perform identification of detected microorganisms?(1) Yes (2) NoQ50In which areas is microbial identification performed?(1) Grade A only (2) Grade A and Grade B (incl. Equivalent areas) (3) All areas including Grade C/D (4) Unknown (5) Other (6) No responseQ51Where is microbial identification performed?(1) In-house (2) Outsourced (external laboratory) (3) Unknown (4) Other (5) No responseQ52Which methods are used for microbial identification?(1) Genetic sequencing (PCR-based) (2) Mass spectrometry (MALDI-TOF MS) (3) Morphological observation (4) Unknown (5) Other (6) No responseEquipment management & digital systemsQ53How are sampling devices managed with respect to movement between areas and sites?(1) Dedicated devices by grade/area; no cross-use (2) Shared across sites within the same grade (3) Shared across grades within the same site (4) Shared across both sites and grades (5) Unknown (6) OtherQ54What is the implementation status of a LIMS for EM data?(1) Implemented (2) Planned/under consideration (3) No plan (4) UnknownQ55Have you performed Environmental Monitoring Performance Qualification (EMPQ)?(1) Yes (performed) (2) No (not performed) (3) Unknown Questionnaire items and response options used in this study. The table lists all questionnaire items (Q01–Q55) with English item wording and corresponding response options; the English text represents translations of the original Japanese questionnaire. Response formats included ordered categorical single-response items (two to five categories), nominal single-choice items, and a small number of items allowing multiple selections. For quantitative analyses, 54 single-response items (Q02–Q55) were used. For items allowing multiple selections (e.g., Q03), when more than one option was selected, the response was recoded into an additional category to enable inclusion in the single-response analytical framework. Data pre-processing and quality assurance Data pre-processing followed a predefined workflow. Ordered categorical single-response items were numerically encoded according to their presentation order to preserve ordinal structure. A total of 54 single-response items were converted into numeric format and used for psychometric evaluation and multivariate analyses. Specifically, when multiple options were selected for a single item, the response was recoded into an additional category (“Multiple selections”). Blank responses or cases in which no applicable option was selected were not treated as missing values. These responses or cases were explicitly coded as a separate “No response” category to distinguish them from valid selections. Internal consistency of the single-response items was assessed using Cronbach's α. Data completeness was confirmed before analysis. Additionally, no missing values were observed across the analytical dataset. Multivariate outlier screening was conducted using Mahalanobis distance calculated from principal component scores, with thresholds based on the χ2 distribution at p < 0.001. No observations exceeded this threshold. Therefore, all respondents and all 54 items were retained for subsequent analyses. Industry classification Respondents selected one of seven manufacturing categories corresponding to their primary operational environment: (Industry 1) sterile drug products (small molecule), (Industry 2) sterile drug products (biologics), (Industry 3) non-sterile drug products, (Industry 4) APIs for sterile use, (Industry 5) non-sterile APIs, (Industry 6) biologics APIs, and. (Industry 7) regenerative medicine products. No respondent selected category (4). Analyses were conducted using the six industry groups represented in the dataset. Multivariate and item-level statistical analyses Principal component analysis (PCA) was applied to the numerically encoded single-response items after standardization to characterize the multivariate structure of EM practices. This aimed to provide a reduced-dimensional representation for downstream analyses. K-means clustering (k = 3) was performed on PCA scores to identify groups of respondents with similar response patterns. Linear discriminant analysis (LDA) was employed to examine the differences among clusters and item-level contributions to cluster separation. Standardized coefficients were used to quantify the relative influence of individual items. To assess whether EM practices differed across manufacturing categories and clusters, permutational multivariate analysis of variance (PERMANOVA) was conducted using Euclidean distance matrices derived from the encoded item responses. Homogeneity of multivariate dispersions was evaluated using betadisper to confirm that PERMANOVA results were not driven by within-group variability. Pairwise PERMANOVA was performed for industry pairs when group sizes permitted valid permutation-based inference. For item-level interpretation, effect sizes (η2) from one-way ANOVA were calculated for each item to quantify the magnitude of between-industry variations. Items with the largest η2 values were examined using response distribution summaries. To characterize the monitoring strategy profiles of the three respondent clusters, a domain-based analysis was performed. Twenty-six questionnaire items were selected to represent six operational domains of environmental monitoring: airborne particle monitoring (Q05, Q06, Q12, Q16, Q17), airborne microbial monitoring (Q18, Q21, Q23, Q27, Q28), surface microbial monitoring (Q36, Q37, Q41, Q42, Q43), media handling and performance testing (Q45, Q46, Q47, Q48), microbial identification (Q49, Q50, Q51, Q52), and equipment management and digital systems (Q53, Q54, Q55). For each item, response distributions across the three clusters were compared using the χ2 test of independence, thereby identifying domains in which cluster-specific monitoring strategies diverged. The modal response and its within-cluster percentage were used to summarize the predominant practice within each cluster for each item. Association between industry group and cluster membership Associations between manufacturing category and cluster membership were evaluated using a χ2 test of independence with simulated p-values. Standardized residuals were examined to identify industry–cluster combinations that were over- or under-represented relative to expected frequencies. Software All statistical analyses were conducted using R version 4.5.2. The psych package was used for reliability analysis, vegan for PERMANOVA and dispersion testing, and base R functions were used for PCA, k-means clustering, LDA, and Mahalanobis distance calculations. Figures were generated from the analytical outputs using standard graphical tools. Results Response rate and data completeness A total of 55 respondents from regulated pharmaceutical and regenerative medicine manufacturing facilities in Japan completed the questionnaire (Fig. 1). Respondents represented six manufacturing categories: sterile small-molecule drug products (I1; n = 19), sterile biologic drug products (I2; n = 9), non-sterile drug products (I3; n = 9), non-sterile active pharmaceutical ingredients (I5; n = 4), biologics APIs (I6; n = 2), and regenerative medicine products (I7; n = 12). Seven product-category options were provided. Moreover, no respondent selected “API for sterile use.” Therefore, six industry groups were included in subsequent analyses. The analytical dataset consisted of 54 questionnaire items (Q02–Q55). After pre-processing and filtering, all 54 items were retained for downstream multivariate analyses. Overview of data acquisition, pre-processing and analysis workflow. (a) Data acquisition: Questionnaire responses on EM practices were collected from regulated facilities manufacturing sterile drug products, non-sterile products, APIs, and regenerative medicine products. (b) Data pre-processing: Responses were screened for quality. Ordered categorical items (2–5 response categories) were numerically encoded. Items were screened for zero variance and perfect correlation; none were identified, and all 54 items were retained. Internal consistency was evaluated via Cronbach's α, and multivariate outliers were assessed using PCA-based Mahalanobis distance. (c) Data analysis: Cleaned data were used to compute industry-wise response profiles, perform PCA and k-means clustering, generate LDA projections, and evaluate between-industry variations using PERMANOVA. Internal consistency of questionnaire items Internal consistency across the analyzed questionnaire items (Q02–Q55; 54 items) was high. Cronbach's α was 0.913, indicating that the survey captured a coherent set of constructs related to EM practices and associated operational decision-making. All 55 respondents provided complete responses for these items, and no missing values were observed across the analytical dataset. To confirm that the multivariate analyses were not unduly influenced by atypical response profiles, data quality checks were conducted before downstream modeling. PCA of the encoded items yielded 14 components with eigenvalues greater than 1 (Kaiser criterion), supporting the multidimensional structure of the questionnaire. Multivariate outlier screening using Mahalanobis distance calculated from the PCA score space identified no respondents exceeding the χ2 threshold at p < 0.001, indicating the absence of extreme multivariate outliers. Together, these results support the inclusion of all respondents and all 54 items in subsequent multivariate analyses. Multivariate structure of environmental monitoring practices PCA was performed on the 54 single-response questionnaire items to summarize the multivariate structure of EM practices across facilities. The first three principal components explained 25.6%, 11.1%, and 7.3% of the total variance. The first ten components accounted for 72.8% of the variance, indicating substantial heterogeneity in EM-related operational profiles (Fig. 2A). Multivariate and industry-wise patterns of encoded response data (a) Principal component analysis: PCA of numerically encoded ordered categorical items illustrates overall variation and grouping tendencies among respondents from different industry sectors. EM: environmental monitoring. (b) Industry-wise response profiles: Radar chart showing the mean encoded response values (range depends on item: 2–5 categories) across items Q02–Q55 for each industry group. Industry-level differences in multivariate response patterns were significant (PERMANOVA: R2 = 0.167, p = 0.001). (c) Pairwise PERMANOVA matrix: Between-industry comparisons identifying significant variations (*p < 0.05). “NA” indicates comparisons where permutation-based testing was not feasible due to small sample size. K-means clustering applied to the PCA score space (k = 3) identified three distinct clusters: Cluster 1 (C1, n = 36; 65.5%), Cluster 2 (C2, n = 13; 23.6%), and Cluster 3 (C3, n = 6; 10.9%). These clusters represent broad differences in overall response patterns rather than variation along a single questionnaire dimension. Industry-wise mean response profiles were generated by averaging encoded item responses within each manufacturing category, revealing clear differences in both response levels and overall patterns among industries (Fig. 2B). Multivariate differences among industry groups were evaluated using PERMANOVA based on Euclidean distances of the encoded item responses. A significant industry effect was observed (R2 = 0.167, p = 0.001), indicating that EM practice profiles differed systematically across manufacturing sectors. Pairwise comparisons further identified specific industry pairs exhibiting significant divergence (Fig. 2C). Cluster membership explained a larger proportion of variance (R2 = 0.331, p = 0.0001), indicating that the derived clusters represent statistically distinct multivariate response structures. To verify that these results were not driven by within-group variability, homogeneity of multivariate dispersions across industries was assessed using betadisper. The test was not significant (p = 0.1110), supporting comparable dispersion across groups. Item-level industry effects and distributions for high-impact items Item-level industry effects were quantified using η2 derived from one-way ANOVA for each of the 54 single-response questionnaire items (Fig. 3A). The five items with the largest effect sizes (Top 5) accounted for 39% of the observed between-industry variation (Fig. 3B). Therefore, industry-level differences in EM practices were driven primarily by a limited subset of questionnaire items. Item-level industry effects and distributions for high-impact items. (a) Top 10 single-response items ranked by industry effect size (η2) derived from one-way ANOVA, indicating items that contributed most strongly to between-industry variation in EM practices. (b) Relative contribution of the top items to the total observed industry effect, expressed as the percentage of cumulative η2 accounted for by each item. (c) Industry-wise response distributions for the five most influential items (Q05, Q41, Q42, Q45, and Q46). Pie charts depict the proportion of response categories within each industry group: sterile drug products (small molecule), sterile drug products (biologics), non-sterile drug products, non-sterile APIs, biologics APIs, and regenerative medicine products. The item with the largest industry effect was Q42 (η2 = 0.4409), which asked whether surface microbial monitoring was implemented in Grade C–equivalent areas (operational ISO 8). The response options were Yes, No, No such area, and Other. Facilities manufacturing sterile drug products—small molecule (Industry 1) and biologics (Industry 2)—reported consistently high implementation, with 89.5% and 88.9% of respondents, respectively, selecting Yes. Biologics API manufacturers (Industry 6) similarly reported uniform implementation (100% Yes). However, non-sterile drug product manufacturers (Industry 3) exhibited a broad distribution of responses. Moreover, only 22.2% selected Yes and a substantial proportion indicated No such area or other response options. Non-sterile API manufacturers (Industry 5) also showed a higher prevalence of No such area responses. Regenerative medicine facilities (Industry 7) largely reported implementation (Yes: 83.3%) but included a small fraction of alternative responses, indicating partial heterogeneity within this sector. The second strongest industry effect was observed for Q46 (η2 = 0.4046), which assessed the frequency of media performance testing. Response options ranged from Each batch/lot to Other. Facilities producing sterile drug products (Industries 1 and 2) most frequently selected intermediate-frequency responses, such as Each purchase. In contrast, non-sterile drug product manufacturers (Industry 3) predominantly selected Before each use (66.7%), indicating higher testing frequency. Regenerative medicine facilities (Industry 7) showed the most pronounced concentration at high frequency, with 83.3% selecting Before each use. Non-sterile API manufacturers (Industry 5) and biologics API manufacturers (Industry 6) exhibited mixed distributions across lower and intermediate categories, reflecting variability in testing practices. A similarly strong industry effect was identified for Q41 (η2 = 0.3773). It asked whether surface microbial monitoring was conducted in Grade B–equivalent areas (operational ISO 7), using the same response structure as Q42. Sterile drug product manufacturers (Industry 1) and biologics API manufacturers (Industry 6) reported predominantly affirmative responses (Yes: 94.7% and 100%, respectively). In contrast, non-sterile drug product manufacturers (Industry 3) and non-sterile API manufacturers (Industry 5) showed a substantial proportion of responses indicating No such area. This reflected the differences in cleanroom classification or operational scope. Regenerative medicine facilities (Industry 7) yielded a mixed pattern, with most respondents selecting Yes, but a minority reporting alternative responses. Two additional items also demonstrated pronounced industry dependence. Q05 (η2 = 0.3412) asked about the timing of airborne particle measurement. The response options included Both at-rest and in-operation, In-operation only, At-rest only, Not clearly distinguished, and Other. Facilities manufacturing sterile drug products (Industries 1 and 2), biologics APIs (Industry 6), and regenerative medicine products (Industry 7) predominantly reported performing measurements in both at-rest and in-operation states. In contrast, non-sterile drug product manufacturers (Industry 3) more frequently selected At-rest only or less clearly defined timing categories. However, non-sterile API manufacturers (Industry 5) exhibited the greatest dispersion across response options. Finally, Q45 (η2 = 0.3252) addressed whether media performance testing was conducted. The response options included Performed in-house, Verified by certificate of analysis (CoA) at receipt, Unknown, and Outsourced. Sterile drug product manufacturers (Industries 1 and 2) and biologics API manufacturers (Industry 6) predominantly reported Performing in-house. In contrast, non-sterile drug product manufacturers (Industry 3), non-sterile API manufacturers (Industry 5), and regenerative medicine facilities (Industry 7) exhibited broader distributions across CoA-based verification, outsourcing, and other approaches, with no single dominant category in Industry 7. These five items indicate that cross-sector differences in EM practices are significantly associated with (i) the scope of surface microbial monitoring across cleanliness grades, (ii) the frequency and management of media performance testing, and (iii) the operational timing of airborne particle monitoring. Fig. 3c shows industry-wise response distributions for these high-impact items. Cluster-wise response profiles and discriminant structures To examine the multivariate structure underlying cluster separation, LDA was applied to the three respondent clusters identified through k-means clustering. Projection onto the first two discriminant functions (LD1 and LD2) revealed clear separation among clusters C1, C2, and C3 (Fig. 4A). LD1 accounted for 98.5% of the between-cluster variance, whereas LD2 accounted for 1.5%, indicating that cluster differentiation was driven almost entirely by a single dominant discriminant axis. Cluster-wise response profiles and discriminant structures (a) LDA scatter plot: Linear discriminant projection (LD1 and LD2) showing separation of clusters, colored by industry group. (b) Cluster-wise profiles: Mean encoded responses for the three clusters (C1, C2, and C3) obtained by k-means clustering on PCA scores (PERMANOVA: R2 = 0.331, p = 0.0001). The clusters comprised 36 respondents in C1 (65.5%), 13 respondents in C2 (23.6%), and 6 respondents in C3 (10.9%). Cluster-wise mean response profiles, obtained by averaging encoded item responses within each cluster, revealed distinct aggregated patterns across the three groups (Fig. 4B). The statistical significance of cluster separation was confirmed by permutational multivariate analysis of variance (PERMANOVA). They demonstrated a strong multivariate effect of cluster membership (R2 = 0.331, p = 0.0001). Therefore, the clusters represent statistically distinguishable response structures among respondents, reflecting systematic differences in reported EM practices rather than random variation. Item contributions to cluster separation and cluster-specific distributions LDA revealed that a limited number of questionnaire items contributed disproportionately to discrimination among the three respondent clusters. Fig. 5A shows the standardized coefficients of the first discriminant function (LD1). LD1 accounted for the vast majority of between-cluster variance, indicating that cluster separation was driven primarily by a single dominant response axis. The ten items with the largest absolute LD1 coefficients explain approximately 54% of the total discriminative contribution (Fig. 5B). Item contributions to cluster separation and cluster-specific distributions. (a) LDA item coefficients: Standardized LDA coefficients indicating the relative contribution of each item to separation among clusters. Red bar shows top five items. (b) Top 10 contributing items: Items with the strongest contribution to LD1-based cluster discrimination. (c) Cluster-wise distributions for representative items: Stacked response categories for selected high-impact items (e.g., Q34, Q27, Q49, Q35, Q31), comparing response patterns among C1, C2, and C3. Among these, several items related to the positioning and justification of microbial sampling heights showed pronounced cluster-dependent response patterns, including Q34, which asked whether the sampling height for settle plate monitoring was defined in standard operating procedures (SOPs).” However, respondents in Cluster 1 (C1) most frequently reported SOP-defined placement (63.9%). Cluster 2 (C2) predominantly selected the same category, indicating that placement was not specified in SOPs but handled inconsistently or variably (76.9%). Cluster 3 (C3) exhibited a consistent mixed pattern, with responses split between SOP-defined placement (33.3%) and non-SOP-defined placement (66.7%). A similar cluster-wise distinction was observed for Q27, which addressed the rationale for selecting the sampling height of active air samplers. In C1, the majority of respondents selected Defined in SOP (72.2%). However, C2 exclusively selected No response (100%), indicating the absence of a clearly articulated rationale. C3 showed an intermediate pattern, with respondents distributed across SOP-defined placement (33.3%), consistent but non-SOP-defined placement (50.0%), and No response (16.7%). Differences in microbial identification practices also contributed to cluster separation. Q49, which asked whether detected microorganisms were identified, showed that C1 overwhelmingly reported performing identification (94.4% Yes). In contrast, C2 exhibited a nearly even split between Yes (53.8%) and No (46.2%). However, C3 predominantly reported not performing identification (66.7% No). These distributions indicate clear cluster-dependent differences in downstream microbiological analysis. Further distinctions were observed for Q35, which examined the rationale for selecting the settle plate sampling height. Respondents in C1 were distributed across multiple predefined rationale categories, including matching to process or working height (33.3%), risk assessment–based placement (16.7%), and defined but unspecified rationales (13.9%), with 36.1% selecting No response. In contrast, C2 exclusively selected No response (100%), whereas C3 showed a predominance of No response (66.7%) with smaller proportions selecting defined rationales. Finally, Q31, which addressed the type of culture media used for the settle plate monitoring, further differentiated clusters. C1 respondents almost uniformly reported using standard media (SCD/TSA; 97.2%). In C2, responses were distributed across standard media (23.1%), alternative media (7.7%), and No response (69.2%). C3 exhibited an intermediate distribution, with two-thirds reporting standard media use (66.7%) and the remainder selecting alternative or unspecified categories. These cluster-specific response distributions indicate that separation among clusters was driven primarily by differences in the degree to which sampling parameters and microbiological procedures were explicitly defined, justified, and reported. Items related to the specification of sampling height, documentation of rationale, performance of microbial identification, and selection of culture media were particularly influential in distinguishing respondent clusters (Fig. 5C). Association between industry groups and cluster membership The distribution of respondent cluster assignments varied across industry groups, with distinct but partially overlapping patterns observed among sectors (Fig. 6A). Respondents from sterile drug product manufacturers—small molecule (Industry 1) and biologics (Industry 2)—were predominantly assigned to Cluster 1. They accounted for 73.7% and 88.9% of respondents in each group, respectively. Biologics API manufacturers (Industry 6) were exclusively assigned to Cluster 1. However, non-sterile drug product manufacturers (Industry 3) showed a more dispersed distribution across clusters, with 22.2% assigned to Cluster 1, 44.4% to Cluster 2, and 33.3% to Cluster 3. Non-sterile API manufacturers (Industry 5) similarly exhibited a heterogeneous pattern, with respondents distributed across all three clusters. Regenerative medicine respondents (Industry 7) were most frequently assigned to Cluster 1 (75.0%), with smaller proportions assigned to Cluster 2 (16.7%) and Cluster 3 (8.3%). Association between industry groups and cluster membership. (a) Distribution of clusters by industry: Proportion of respondents in each cluster (C1–C3) across industry groups (I1–I7). (b) Standardized residuals from the χ2 test: Heatmap showing deviations between observed and expected industry–cluster combinations. Positive values indicate over-representation; negative values indicate under-representation. The association between industry group and cluster membership was evaluated using a χ2 test of independence (Fig. 6B). The overall test did not reach conventional statistical significance (χ2 = 15.95, p = 0.0935). Thus, the effect size was moderate (Cramér's V = 0.381), suggesting a non-negligible association between industry category and response cluster. Examination of standardized residuals revealed specific industry–cluster combinations that deviated significantly from expected frequencies. Non-sterile drug product manufacturers (Industry 3) were significantly under-represented in Cluster 1 and over-represented in Cluster 3, indicating a cluster allocation pattern distinct from that of sterile manufacturing sectors. No other industry–cluster combinations exceeded the predefined thresholds for significant deviation. These results demonstrate that cluster membership was not uniformly distributed across industry groups and that specific industry–cluster combinations exhibited statistically detectable deviations from expected frequencies (Table 2). Table 2Domain-based monitoring strategy profiles of respondent clusters.Table 2ItemDescriptionC1 (n = 36)C2 (n = 13)C3 (n = 6)Sig.Airborne particle monitoringQ05Measurement timingBoth at-rest & in-opBoth at-rest & in-opAt-rest only***86%46%33%Q06Sampling point rationaleISO + risk combinedISO + risk combinedUnknown; in SOP***44%46%67%Q12Plan review frequencyNot changed unless issueNot changed unless issueNot changed unless issue**53%54%67%Q16Sampling height definitionDefined in SOPConsistent, not in SOPConsistent, not in SOP***81%85%50%Q17Sampling height rationaleProcess/working heightNo responseNo response***47%85%67%Airborne microbial monitoringQ18Sampling methodBoth active & settleActive onlyBoth active & settle***97%77%50%Q21Measurement timingBoth at-rest & in-opNo responseIn-operation only***56%100%33%Q23Delay time settingSetNo responseNot set***53%92%67%Q27Sampling height definitionDefined in SOPNo responseConsistent, not in SOP***72%100%50%Q28Sampling height rationaleProcess/working heightNo responseNo response***39%100%67%Surface microbial monitoringQ36Sampling methodContact platesContact platesContact platesns86%69%50%Q37Sampling point rationaleRisk assessmentRisk assessmentRationale unknown***83%62%33%Q41Implementation (Grade B)YesYesNo such area**78%62%33%Q42Implementation (Grade C)YesYesNo such area***86%62%33%Q43Implementation (Grade D)YesYesNo such area*53%77%33%Media handling and performance testingQ45Testing approachIn-houseIn-houseVerified by CoA**72%54%33%Q46Testing frequencyEach purchaseEach purchaseBefore each usens36%46%100%Q47Guideline basisJPJPUnknownns72%77%50%Q48Acceptance criteriaQuantified ref. strainsMean values at adoptionUnknown*53%38%83%Microbial identificationQ49Identification performedYesYesNo***94%54%67%Q50Target areasGrade A and BNo responseNo response**58%46%67%Q51Testing locationIn-houseNo responseNo response***56%46%67%Q52Identification methodGenetic sequencingNo responseNo response*33%46%67%Equipment management and digital systemsQ53Device managementShared across gradesDedicated by gradeShared across gradesns39%31%33%Q54LIMS implementationImplementedImplementedUnknownns50%38%50%Q55EMPQ performedUnknownUnknownUnknownns42%46%83% Domain-based monitoring strategy profiles of respondent clusters. Domain-based monitoring strategy profiles of respondent clusters To characterize the monitoring strategies represented by each cluster, response distributions for 26 items spanning six operational domains were compared across C1, C2, and C3 (Table 2). Of the 26 items examined, 20 showed statistically significant differences across clusters (p < 0.05), whereas six did not, indicating that cluster-level variation was concentrated in specific operational domains. In the airborne particle monitoring domain, all five items differed significantly across clusters. Cluster 1 respondents predominantly reported SOP-defined sampling heights (Q16: 81%) with explicit rationales based on process or working height (Q17: 47%), and most performed measurements in both at-rest and in-operation states (Q05: 86%). Cluster 2 respondents reported consistent but non-SOP-defined sampling heights (Q16: 85%) and lacked documented rationales for height selection (Q17: 85% no response). Cluster 3 exhibited the most dispersed response patterns, as indicated by at-rest-only measurements (Q05: 33%) and undocumented sampling point rationales (Q06: 67% unknown, in SOP). In the airborne microbial monitoring domain, all five items differed significantly. Cluster 1 respondents reported comprehensive monitoring using both active and settle plate methods (Q18: 97%), with SOP-defined sampling heights (Q27: 72%) and established delay time settings (Q23: 53%). In contrast, Cluster 2 was characterized by a near-complete absence of documented parameters: 100% of respondents selected “no response” for sampling height definition (Q27), sampling height rationale (Q28), and measurement timing (Q21). Cluster 3 exhibited intermediate patterns, with both active and passive methods used (Q18: 50%) but without established delay times (Q23: 67% not set). In the surface microbial monitoring domain, four of five items showed significant cluster differences. Sampling method (Q36: contact plates) did not differ significantly, indicating a shared methodological basis. However, monitoring coverage differed: Cluster 1 reported high implementation rates in Grade B (Q41: 78%) and Grade C areas (Q42: 86%), Cluster 2 showed moderate implementation (Q41: 62%, Q42: 62%), and Cluster 3 reported low implementation with a high proportion of “no such area” responses (Q41 and Q42: 33% each), reflecting limited cleanroom scope. In the media handling and performance testing domain, two of four items showed significant differences. Performance testing was predominantly conducted in-house in Clusters 1 and 2 (Q45: 72% and 54%, respectively), whereas Cluster 3 relied more frequently on certificate-of-analysis verification (Q45: 33%). Testing frequency (Q46) and guideline basis (Q47) did not differ significantly across clusters. However, acceptance criteria (Q48) differed: Cluster 1 used quantified reference strains (53%), whereas Cluster 3 predominantly selected “unknown” (83%). In the microbial identification domain, all four items differed significantly. Cluster 1 reported routine identification of detected microorganisms (Q49: 94%), targeted primarily at Grade A and B areas (Q50: 58%), conducted in-house (Q51: 56%), and using genetic sequencing methods (Q52: 33%). Clusters 2 and 3 showed lower identification rates (Q49: 54% and 33%, respectively), with high proportions of “no response” for target areas, testing location, and identification method. In the equipment management and digital systems domain, none of the three items showed significant cluster differences. LIMS implementation (Q54), device management practices (Q53), and EMPQ performance (Q55) were uniformly distributed across clusters. The three clusters represented distinct monitoring strategy profiles that differed across five of the six operational domains. The domains with the largest cluster differences were airborne particle and microbial monitoring, surface microbial monitoring, and microbial identification — all of which relate to operational decision-making and documentation practices. Equipment and systems infrastructure did not differ across clusters. Discussion This survey provides a quantitative overview of how environmental monitoring (EM) practices are implemented across regulated manufacturing sectors in Japan and places regenerative medicine manufacturing within that broader operational landscape. The results showed that EM practices differed across sectors, but these differences were not explained by industry category alone. Rather, the observed variation appeared to reflect a combination of manufacturing workflow, cleanroom design, monitoring scope, and local quality-system implementation. For regenerative medicine manufacturing, this finding is particularly important because contamination control strategies are developed considering common regulatory requirements — including maintaining appropriate cleanliness classifications, routine environmental monitoring, and validated cleaning and disinfection procedures — yet must be adapted to highly diverse operational settings that differ in product characteristics, required cell counts, container types, and aseptic handling procedures. Across manufacturing sectors, respondents from sterile drug product manufacturing environments generally reported more standardized approaches to EM. This included clearly defined rationales for sampling decisions and structured assignment of monitoring responsibilities, consistent with the grade-based monitoring framework in which EM is tightly linked to sterility assurance [4]. However, respondents from non-sterile drug product and active pharmaceutical ingredient manufacturing more frequently reported diverse or less uniform practices. This was consistent with the functional-area-based monitoring approach described above, in which sampling scope and frequency are adapted to operational risk rather than defined by cleanliness grade. However, the cluster analysis demonstrated that such patterns were not exclusive to specific industries, highlighting substantial within-industry heterogeneity in EM implementation. Respondents associated with regenerative medicine manufacturing exhibited heterogeneous EM practices rather than a single, uniform profile. Regenerative medicine products are manufactured under regulations specific to cell and tissue-based products and their manufacturing may involve the use of starting materials with inherent bioburden [[12], [13], [14]]. In addition, given the diversity of manufacturing formats—ranging from open cleanroom operations requiring extensive operator interaction to closed or semi-closed automated processing systems—contamination risks and the required monitoring intensity vary substantially even within this single product category. Factors such as the required cell count, the type of container used for processing, and the degree of aseptic manipulation further influence how EM programs are designed. Accordingly, these results do not support a simple one-to-one correspondence between regenerative medicine classification and specific EM practice patterns. The domain-based cluster profiling further clarified the nature of the abovementioned variations. Cluster 1 exhibited a mature, well-documented monitoring strategy across all operational domains: sampling parameters were defined in SOPs, rationales for sampling decisions were explicitly recorded, comprehensive monitoring methods were employed (both active and settle plate sampling for airborne microorganisms; surface monitoring in Grade B and C areas), microbial identification was routinely performed in-house using molecular methods, and media performance testing was conducted internally with quantified acceptance criteria. Cluster 2 was characterized by absence of documented rationales and parameters, particularly in airborne monitoring domains where 100% of respondents lacked documented sampling heights, timing, and rationale. Despite this documentation gap, Cluster 2 respondents reported moderate implementation of surface monitoring and in-house media testing, suggesting that monitoring activities were performed without formal quality-system documentation. Cluster 3 exhibited the most limited monitoring scope: surface monitoring was implemented at low rates, microbial identification was largely absent, and acceptance criteria for media testing were predominantly unknown. However, Cluster 3 uniformly reported the highest media testing frequency (100% before each use), indicating that limited-scope programs may compensate through intensified testing in selected areas. Critically, the equipment management and digital systems domain showed no significant differences across clusters. LIMS implementation, device management practices, and EMPQ performance were similarly distributed regardless of cluster membership. This finding highlights operational decision-making and quality-system maturity as the distinguishing factors between clusters rather than the differences in available infrastructure. The respondents from all six manufacturing sectors were distributed across the three clusters, rather than segregated into product-specific groups. This observation indicates that monitoring strategies are shaped by organizational structure, facility design, and institutional approaches to quality management rather than by product category alone [7,[18], [19], [20], [21], [22]]. The results demonstrate that EM practices under shared regulatory requirements exhibit both cross-sector differences and pronounced variability within sectors, and that facility- and organization-level factors — such as documentation practices, operational responsibility allocation, and historical implementation choices — are crucial when evaluating EM programs. These observations are consistent with risk-based principles emphasized in contemporary regulatory and technical guidance [3,4,11] and underscore that effective EM systems must be calibrated to realistic contamination risks and operational constraints rather than defined solely by product category. This study has several limitations. First, the survey was conducted at the respondent level, and multiple respondents may have originated from the same manufacturing site. Because facility identifiers were not collected, responses could not be aggregated at the site level. Therefore, the results reflect individual perceptions and reported practices rather than formally harmonized, facility-wide EM programs. Second, the number of respondents in some industry categories, particularly biologics API and non-sterile API manufacturing, was limited. This imbalance may have reduced statistical power for detecting industry–cluster associations and contributed to non-significant global χ2 results despite moderate effect sizes. Third, the analysis focused on reported EM implementation patterns and did not directly assess relationships with objective outcomes such as environmental microbial counts, deviation rates, or product contamination events. Previous studies have shown that environmental monitoring results do not always correlate directly with product bioburden, reflecting the inherent variability of sampling and the complexity of contamination pathways [23]. These findings must be interpreted as descriptive of EM implementation patterns rather than as indicators of EM effectiveness. Fourth, all questionnaire items were numerically encoded and analyzed using methods based on Euclidean distance (PCA, PERMANOVA) and linear projections (LDA). This approach assumes at least ordinal measurement properties across all items. However, a subset of items, including Q02 (department responsible for EM), represents nominal classifications for which the assigned numeric ordering is arbitrary. This constitutes a known limitation, as the application of metric-based multivariate methods to nominal data may introduce systematic bias in distance calculations. Importantly, the items identified as primary contributors to between-group variation (Q05, Q41, Q42, Q45, Q46) all have ordinal or dichotomous response formats, which mitigates the potential impact of this limitation on the principal findings. Nonetheless, future cross-sectional studies of EM practices should consider explicitly distinguishing ordinal and nominal items at the design stage and employing distance measures appropriate for mixed measurement scales. Conclusion This study provided a systematic, data-driven characterization of how environmental monitoring practices are implemented across regulated manufacturing sectors in Japan. By applying multivariate statistical approaches, it identified key questionnaire items and response patterns that account for a substantial proportion of between-industry and between-cluster variation. The results demonstrated that, even under shared regulatory requirements for environmental control across manufacturing sectors, EM practices diverge according to manufacturing context, operational workflows, and organizational approaches to quality management. Regenerative medicine facilities showed partial alignment with more standardized sterile manufacturing environments but also indicated substantial heterogeneity in implementation. Differences in surface microbial monitoring coverage, media performance testing practices, and monitoring timing were major contributors to cross-sector variability. These findings offer a quantitative foundation for understanding current EM implementation patterns and are expected to support future efforts to refine risk-based environmental monitoring strategies in regenerative medicine and other diverse manufacturing environments. Author contributions Mitsuru Mizuno: Conceptualization, methodology, numerical encoding of questionnaire responses, formal analysis, statistical modeling, data interpretation, visualization, writing–original draft, Writing–review & editing. Tomoko Oono: Data curation, numerical encoding of questionnaire responses, visualization, writing – review & editing. Kana Takahashi: Questionnaire design, data acquisition, writing–review & editing. Kazuko Tsujimoto: Questionnaire design, data acquisition, writing–review & editing. Natsuki Tanigawa: Questionnaire design, data acquisition, writing–review & editing. Yoshinobu Nakamura: Questionnaire design, data acquisition, writing–review & editing. Hiroaki Nakamura: Conceptualization, questionnaire design, data acquisition, data interpretation, writing – review & editing. Kenji Fujii: Questionnaire design, data acquisition, writing–review & editing. Ichiro Sekiya: Supervision, Writing – review & editing. All authors reviewed and approved the final manuscript. Data and code availability The R analysis code used to generate the statistical results reported in this study is publicly available at the Open Science Framework (https://doi.org/10.17605/OSF.IO/V546H). The survey response data contain individual-level information from a limited number of respondents at regulated pharmaceutical and regenerative medicine manufacturing facilities. Due to the small sample size and the potential for re-identification of participating facilities, the raw data are not publicly deposited. Data may be made available upon reasonable request to the corresponding author. Funding This research was funded by a joint research grant from the Earth Environment Service Corporation (Japan). Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Mitsuru Mizuno reports financial support was provided by Earth Environmental Service Co., Ltd. Acknowledgements We thank Chiaki Okumura, Asuka Asami, and Hisako Katano for their assistance in managing the laboratory. Footnotes Peer review under responsibility of the Japanese Society for Regenerative Medicine. References Associated Data Data Availability Statement The R analysis code used to generate the statistical results reported in this study is publicly available at the Open Science Framework (https://doi.org/10.17605/OSF.IO/V546H). The survey response data contain individual-level information from a limited number of respondents at regulated pharmaceutical and regenerative medicine manufacturing facilities. Due to the small sample size and the potential for re-identification of participating facilities, the raw data are not publicly deposited. Data may be made available upon reasonable request to the corresponding author.