Challenging Patterns of Fine‐Scale Genetic Structure in Galicia and East‐to‐West Genetic Divergence Across the Iberian Peninsula—When Technical Batches Rewrite History

ABSTRACT

The influential 2019 study by Bycroft et al. reported remarkably pronounced ultrafine‐scale genetic structure within a small region of Galicia (northwestern Iberia). Using fineSTRUCTURE clustering of ChromoPainter coancestry profiles, the authors identified multiple internally coherent clusters characterized by high levels of within‐group haplotype sharing and reduced sharing with neighboring populations. On a broader national scale, they also reported a marked East–West genetic differentiation and North–South homogeneity across the Iberian Peninsula, patterns which they largely attributed to historical population movements associated with the gradual territorial expansion of Christian kingdoms between the 8th and 15th centuries following the initial Islamic conquest. Several subsequent studies have adopted similar interpretations, but all relied on the same genetic resource, the EPICOLON Phase I cohort. Following a detailed reexamination of Bycroft et al., we identify multiple methodological issues, most notably signals consistent with unaccounted batch effects in EPICOLON, some of which were recognized by the authors in earlier studies. In contrast, we highlight recent analyses based on independent genomic datasets that consistently report high levels of genetic homogeneity within Galicia and across central and southern Iberia, with no evidence supporting a pronounced East–West genetic divide across Iberia. In addition, recent studies have documented North African demographic influences in Iberia, particularly in Galicia, predating the period of Islamic rule and therefore challenging the historical interpretation proposed by Bycroft et al. We also show that while fineSTRUCTURE is a powerful tool for detecting subtle population stratification, it may be highly sensitive to technical confounders, which can magnify or distort inferred clustering patterns. This study demonstrates that technical artifacts can strongly affect fine‐scale population structure inferred by fineSTRUCTURE, emphasizing the need for rigorous quality control and cautious interpretation of highly resolved genetic clusters.

Findings reported by Bycroft et al. [1] for the Iberian Peninsula, including ultra‐fine population structure in Galicia, East–West genetic differentiation, and North African ancestry associated with the ‘Reconquista’ period, may be influenced by technical artifacts and sampling limitations within the EPICOLON Phase I dataset. These potential biases could contribute to discrepancies with historical records, Iberian geographic expectations, and previous scientific evidence, highlighting the need to reassess EPICOLON‐derived inferences using independent cohorts and robust analytical frameworks.

1. Background

Bycroft et al. [1] conducted an in‐depth genetic analysis of the Iberian Peninsula, revealing significant fine‐scale population structure and deep genetic stratification. Utilizing the fineSTRUCTURE method and the GWAS Spanish dataset known as EPICOLON Phase I (see details in Supporting Information), they identified 145 distinct clusters, with a particular emphasis on what they considered highly reliable cluster assignments, especially in geographically localized regions such as Galicia (located in the northwestern corner of the Peninsula) and the Basque Country (in the North). The study also highlighted a major axis of genetic differentiation running East to West across the peninsula and minor genetic differences in the North–South direction across Iberia, patterns that the authors attributed to population movements during the Reconquista, arguing that migrations from northern Christian kingdoms into newly conquered territories generated genetic boundaries that closely reflect historical regional divisions. The study has received significant academic attention and widespread discussion on social media platforms.

Following these findings, both Pimenta et al. [4] and Ferreiro et al. [5] reported Galicia as a genetically distinctive region within the Iberian Peninsula, attributing this pattern to a combination of historical, cultural, linguistic, and demographic factors. Pimenta et al. [4] emphasized the close genetic relationship between Galicia and Portugal and proposed that genetic structure across Iberia reflects historical political and cultural boundaries. Ferreiro et al. [5] similarly identified Galicia as genetically differentiated and argued that historical isolation, language persistence, and demographic processes contributed to the observed patterns. However, both studies relied on the same EPICOLON dataset. Consequently, their findings should not be considered independent confirmations of the reported patterns, but rather as interpretations derived from essentially the same underlying genetic resource. An additional indication of the influence of findings by Bycroft et al. [1], likely magnified by its publication in a high‐impact journal, comes from Garcia‐Calleja et al. [6], who used an independent dataset but supported the same hypothesis without presenting new evidence (Supporting Information).

We have thoroughly examined the study of Bycroft et al. [1] and identified numerous significant anomalies, most likely arising from technical artifacts and other methodological issues, that may account for most of the patterns reported and their conclusions. Given that the same dataset was used in other studies addressing similar historical questions [4, 5], there is a clear risk that all of them may also be affected by the same technical artifacts, potentially reinforcing historical interpretations derived from the defective dataset.

While our main focus is a close inspection of the EPICOLON dataset and the pioneering study by Bycroft et al. [1], we also highlight that recent analyses of the Galician population, using independent genomic datasets [2], reveal high genetic homogeneity within Galicia, contrasting with the pronounced stratification reported by [1]. Additionally, analyses by Pardo‐Seco et al. [2] have recently reveal genetic continuity along an East–West axis across the southern and central Iberian Peninsula, and a lack of homogeneity along the North–South axis, again differing from the patterns reported in [1]. Finally, we provide a proof‐of‐concept showing that, while fineSTRUCTURE is a powerful tool for detecting fine‐scale population stratification based on haplotype‐sharing patterns, its results may be sensitive to technical artifacts capable of altering coancestry profiles and generating spurious clusters.

2. Results

2.1. The EPICOLON Phase I Dataset and Technical Issues

Bycroft et al. [1] used genomic data from Fernández‐Rozadilla et al. [7], which were part of the EPICOLON Phase I colorectal cancer genome‐wide association study (GWAS). The EPICOLON cohort consisted of 1548 donors, including 881 individuals with colorectal adenocarcinoma and 667 controls, with ~68% recruited from five hospitals in Catalonia (northeast of Spain) and Hospital do Meixoeiro in Vigo, Galicia (northwest of Spain). However, information about the recruiting centers remains ambiguous. This issue is particularly sensitive, given that the main argumentation in the present report concerns potential batch effects (Supporting Information).

As explained in their manuscript, 1413 samples were retained after quality control and used to generate the main fineSTRUCTURE dendrogram. Of these, 726 individuals whose four grandparents were born within 80 km of their birthplace were retained for the final geographic analysis. These individuals were assigned to clusters inferred from the full set of 1413 samples. Consequently, the interpretation of the results relies on the assumption that the inclusion of the remaining samples, for which equivalent geographic constraints were not applied, did not materially influence the clustering structure inferred from the full dataset. One of the most notable findings was the unexpectedly pronounced ultra‐fine‐scale genetic structure reported within Galicia, suggesting its genetic uniqueness relative to other regions of Spain.

However, Bycroft et al. [1] did not address key technical issues, particularly batch effects related to the hospital of origin; concerns that had already been raised in connection with the EPICOLON dataset (Supporting Information). Thus, Fernández‐Rozadilla's PhD dissertation [8], one of the coauthors in [1], along with other previous publications using the EPICOLON dataset [7, 9], highlighted the risk of population stratification arising from batch effects stemming from variations in sample collection and DNA extraction procedures. As mentioned in Fernández‐Rozadilla et al. [7], the EPICOLON dataset was processed in 23 separate batches, each containing between 16 and 99 samples according to hospital of origin, introducing the potential for batch‐related distortions (Supporting Information).

As we argue below (and in Supporting Information), batch effects in EPICOLON may offer a more parsimonious explanation for their extraordinary findings. This omission is particularly significant given the high sensitivity of fineSTRUCTURE for detecting subtle genetic stratification. As we demonstrate below, this tool can be highly sensitive to both genuine and artifactual patterns in genetic data. Moreover, the peer review process failed to identify these technical limitations (reports are publicly available on the journal's website).

2.2. Main Anomalies in the Bycroft et al. (2019) Study

A comprehensive description of the anomalies discussed in this section is provided in the Supporting Information. Figures S1–S11 in this document are intended to support the interpretation of the methodological and technical issues discussed in sections 2.2, 2.3 and 2.4. Several of these figures contain edited versions of the maps published by Bycroft et al. [1] intended to facilitate interpretation and highlight key features.

2.2.1. Artificial Ultra‐Fine‐Scale Stratification in Pontevedra, Galicia

Bycroft et al. [1] reported extensive fine‐scale genetic structure within Spain, with an unexpectedly high level in the province of Pontevedra (in the south‐western corner of Galicia), where they identified an extremely dense concentration of genetic clusters, particularly around the port city of Vigo and its surrounding southern and eastern areas. The authors stated that fineSTRUCTURE inferred 145 distinct clusters in the entire Spanish (EPICOLON) dataset, noting that “by far the strongest sub‐structure is seen within a single province in Galicia, Pontevedra, which contains almost half of the inferred clusters in all of Spain”. To put this in context, the area referenced by the authors is a small region within Pontevedra, corresponding roughly to a compact, rectangular area approximately 20 km North–South, and 60 km East–West. The authors attributed this fine‐scale genetic differentiation to geographic features such as hills and river valleys, despite the absence of relevant barriers in the region, and argued that genetic structure could be detected at scales as small as 10 km.

The clustering process employed by the authors is problematic, as it was based on the full cohort despite the absence of geographic origin data for approximately half of the samples (726 out of 1,413). Consequently, geographic labels were assigned to clusters inferred from haplotype‐sharing similarity, even though sample provenance metadata were incomplete (Supporting Information). The subsequent analyses rely heavily on this initial clustering step and therefore inherit these limitations.

Upon closer examination, the results appear even more extraordinary than those described by the authors. A meticulous inspection of their complete fineSTRUCTURE Iberian tree (Figure 1A) reveals that more than 72% of the inferred clusters and internal splits (> 105 of 145) are concentrated solely in Pontevedra, far exceeding the “about half” reported by the authors. This pattern indicates an exceptionally high degree of haplotype‐sharing differentiation within this very small area. In contrast, most other regions of continental Spain are represented by only one to three branches; for example, the entire Basque Country has just seven. Most strikingly, a count of the individuals shown in the authors' maps (their figure 1B and the zoomed‐in genetic map in their figure 3A) suggests that there are more inferred clusters than sampled individuals fulfilling the criterion of having four grandparents born within the same Galician region (Supporting Information). This discrepancy calls for closer scrutiny of the robustness of the EPICOLON dataset used in their genetic clustering analyses and raises significant concerns about potential technical and sampling‐related biases in the interpretation and representation of their findings.

FIGURE 1.

Comparison of the contrasting hierarchical clustering structures inferred from coancestry matrices reported by Bycroft et al. [1] and Pardo‐Seco et al. [2] for the Galician population. (A) Full tree reported by Bycroft et al. [1] in their figure 1A, highlighting the extreme genetic stratification described for the small region in the Galician province of Pontevedra. (B) Modified version of the clustering tree reported by Pardo‐Seco et al. [2] for Galicians using WGS data, indicating the presence of only five clusters, with most of the samples concentrated in the ‘Main’ cluster. In both cases, the fineSTRUCTURE tool was used to infer clusters. A heatmap of pairwise Total Variation Distance (TVD) for the GAL dataset is provided in Figure S8.

The high degree of localized genetic differentiation observed in Pontevedra, particularly at resolutions under 10 km, may be influenced by technical issues in the EPICOLON dataset. Previous reports of the authors have noted chemical and procedural problems in sample processing that could have artificially induced fine‐scale stratification [7]. A critical issue is the potential sampling bias, as all Galician samples were collected more than a decade before the publication of the study from a single institution, Hospital do Meixoeiro (Vigo city), originally for biomedical research purposes. This hospital is in a highly urbanized area with well‐established transportation links. This characteristic of the region contradicts the authors' assertion of extreme genetic isolation, as such urban and economic connectivity should theoretically promote genetic exchange rather than isolation. The interconnected nature of areas like Vigo (a historically significant port city located in the core of this highly stratified region), Tui, A Guarda, Ponteareas, and Tomiño, with populations exceeding 10,000–20,000, further undermines the claim of extreme genetic isolation of this key area. Moreover, the authors' assertion that Galicia's genetic structure is primarily shaped by rural living and linguistic uniqueness is difficult to reconcile with observations from other linguistically distinct regions of Spain, such as Catalonia and the Basque Country (speaking a pre‐Indo‐European language), which do not exhibit comparable genetic fragmentation. In addition, the widespread use of a common language throughout Galicia (Galician) would not be expected to promote genetic differentiation within the region. Additionally, the region's long history as an important urban and religious center, including cities like Tui, also located at the epicenter of the area identified as highly genetically stratified, challenges the portrayal of Galicia as a genetically isolated area. Tui, for example, was one of the most important Christian centers during the early medieval territorial expansion of Christian kingdoms, making the claim of genetic isolation at that time even more dubious (Supporting Information).

Previous research by Fernández‐Rozadilla et al. [7, 8, 9] documented technical biases, particularly related to sample processing, which, in our view, provide a more parsimonious explanation for these unusual patterns than do genuine historical or geographical factors. Accordingly, the extraordinary findings raise concerns about methodological flaws, suggesting that the observed genetic patterns may reflect technical artifacts rather than genuine historical or geographical factors. In addition, the marked contrast between the highly structured results from Bycroft et al. [1] and the very limited genetic differentiation observed in the high‐resolution Whole Genome Sequencing (WGS) data generated within the GALOMICS (GAL) project [2], further supports this hypothesis. This discrepancy warrants careful consideration (see below and Supporting Information).

2.2.2. Sample Quality Issues and National‐Level Genetic Patterns

Bycroft et al. [1] identified national‐scale genetic patterns that may also stem from technical artifacts rather than genuine population structure. For instance, they described Galicia as exhibiting the earliest genetic split from the rest of Spain. This result appears to contradict earlier studies suggesting that the region was a genetic cul‐de‐sac shaped by post‐glacial migrations rather than continuous habitation [10] (see a more detail argumentation in Supporting Information). While the Cantabrian Mountain range, which runs from East to West parallel to the northern coast of the Iberian Peninsula, has historically acted as a barrier to gene flow toward the southern part of the peninsula, the authors observed greater genetic continuity across it than within the relatively small Pontevedra region, where no significant geographical barriers are present. This inconsistency raises concerns about the validity of their inferred North–South genetic homogeneity and East–West genetic stratification across the Iberian Peninsula.

Two major national‐scale patterns in the Spanish maps provided by Bycroft et al. [1] are particularly problematic (Supporting Information). The considerations discussed below should be viewed as interpretative and exploratory, as their validity can only be formally assessed through examination of the original data and metadata. First, their “Central” cluster (yellow band in e.g., Figures S4–S6) may closely align with a batch of samples derived from their National DNA bank (NDNAB) dataset, all of which could have been processed in the same laboratory. The uniform appearance of this cluster, extending from North to South and broadly distributed across the Iberian Peninsula, suggests that sample processing, rather than true genetic similarity, may have contributed to its distinct grouping. Second, their “West” cluster (red band in e.g., Figures S4–S6) exhibits greater haplotype‐sharing similarity with samples from Catalonia and the Balearic Islands (distantly located in the easternmost part of the Iberian Peninsula and the Mediterranean Sea, respectively) than with geographically adjacent regions, such as Cantabria, Navarra, and the Basque Country (see their fineSTRUCTURE tree of figure 1A). This unexpected pattern raises the possibility of technical bias rather than reflecting genuine historical genetic connections. For instance, if many of these samples assigned to the “West” cluster were processed at the same facility, such as the Hospital de Donostia (Basque Country), shared technical factors could have introduced systematic biases that contributed to the anomalous grouping. Notably, some members embedded within the ‘red band’ cluster belong to typical Basque clusters (their figure 1B; Supporting Information). Collectively, these issues cast doubt on Bycroft et al.'s [1] claim of an East–West genetic divide in Iberia.

Close inspection of the study by Bycroft et al. [1] revealed further notable inconsistencies that are not a matter of interpretation but are directly observable from their published descriptions and figures. The authors included 107 Portuguese individuals for whom no biogeographic information was available (as stated by the authors). These samples were displayed as being geographically randomly distributed across mainland Portugal, although the authors did not elaborate on the rationale underlying this decision. These samples received the same color in their maps as the ‘Main’ Galician cluster. This arbitrary decision artificially suggests a smooth, longitudinal, homogeneous genetic pattern between Northern Galicia and Southern Portugal, thereby making their purported East‐to‐West pattern across the entire Peninsula appear much more convincing. Also problematic, this view completely obscures the very distinctive genetic features they describe for the Pontevedra area, which, according to the authors, contains almost half of the inferred clusters in all of Spain. Furthermore, although they identified 145 genetic clusters in Spain, their analyses assigned only a single cluster to all of Portugal.

Finally, the inclusion of Portuguese samples caused Galicia to be split into two clear coancestry‐based non‐sister branches within the fineSTRUCTURE dendrogram. This pattern is difficult to reconcile with an explanation based on recent isolation alone. Rather than reflecting recent demographic separation, the results imply a strong and persistent differentiation in haplotype‐sharing profiles within Galicia. This would require the long‐term historical coexistence of two markedly dissimilar and minimally admixed coancestry profiles within a geographically very small and historically well‐connected region within Galicia and with Portugal (Supporting Information).

2.2.3. Sampling Bias and Its Impact on Results

Although not explicitly stated, the maps in Bycroft et al. [1] exhibit a notable urban sampling bias, with most individuals seemingly recruited from major provincial capitals or large hospitals in cities such as Barcelona, Lleida, Tarragona, Zaragoza, and Oviedo (Supporting Information). The spatial distribution of their samples suggests a particularly high concentration of urban participants in regions like Catalonia, especially from cities like Barcelona, Lleida, and Tarragona, whereas regions such as the Basque Country and Galicia appear to have more rural sampling patterns (Supporting Information). This sampling bias is important, as individuals from urban areas are generally more likely to exhibit diverse ancestral origins compared to those from rural settings. However, the study does not address this issue, which may lead to misinterpretation of the observed genetic patterns. If sampling strategies varied geographically, favoring urban centers in some regions and rural areas in others, this could artificially inflate the apparent genetic structure in certain areas while obscuring it in others. For example, while we suspect that the extreme genetic fragmentation observed in Galicia is most likely due to technical issues, it could also be plausible that an overrepresentation of individuals from specific rural locations, rather than true population structure, may have contributed to the observed patterns. Since Bycroft et al. [1] did not provide detailed information on the geographic origins of their samples, there is ambiguity regarding whether donors were from capital cities, smaller towns, or rural areas. Moreover, in Spain, province names often coincide with the names of capital cities, meaning a sample labeled “Pontevedra” could refer either to the city or the entire province (Supporting Information). It is therefore conceivable that the geographic resolution for most of the biogeographically identified donors was limited to the provincial level, forcing the authors to assign locations based on the capital of the corresponding provinces.

Peer reviewers raised concerns regarding these biases, questioning whether observed differences in population structure reflect true genetic variation or sampling‐related biases (Supporting Information). Despite acknowledging this possibility, Bycroft et al. [1] did not provide sufficient transparency about their sampling methods, limiting the reliability of their conclusions (Supporting Information).

2.2.4. Issues With Basque (And Other) Genetic Assignments

Bycroft et al.'s [1] dataset includes individuals assigned to Basque genetic clusters despite residing outside the Basque Country, such as in Extremadura in the South (Supporting Information). For this classification to be accurate, these individuals would need to have multiple generations of Basque ancestry without admixture with locals, despite living far outside the Basque region, as the authors state that all four grandparents of each donor were born within 80 km of their average birthplace. A more plausible explanation, however, is that these individuals were processed in the same hospital as other Basque samples, leading to systematic technical biases. To clarify this issue, it would be essential to determine whether these Basque‐assigned individuals (and all donors in EPICOLON) were sampled and processed in the same batch as other Basque donors (e.g., Hospital de Donostia). If this is the case, it would suggest that their fineSTRUCTURE genetic clustering is an artifact of data processing rather than a reflection of genuine ancestry.

Similar examples can be observed in other parts of the map and with other regions: for instance, one individual seemingly belonging to a very ‘genuine’ Galician clade (highlighted in their zoomed‐in maps of figure 3A) appears in the South of the Basque Country (Supporting Information). In this case, we hypothesize that the sample, originally from the Basque Country, may have artificially acquired a Galician cluster designation simply due to being processed in the Galician batch of Hospital do Meixoeiro. The Supporting Information provides additional examples of these counterintuitive, biogeographically ‘misallocated’ genomes.

2.3. Most Recent Analyses Contradict Bycroft et al. (2019) Findings

The recent analysis carried out by Pardo‐Seco et al. [2], using the highest level of genetic resolution possible of the GAL WGS dataset (including town‐level geographic resolution), yielded results that differ substantially from those reported by [1], while taking into account micro‐regional differences in the sampling areas. Pardo‐Seco et al. [2] identified 10 terminal branch clusters in Galicia (compared to Bycroft et al.'s > 105 out of 145 in Iberia), with only five having at least four individuals; Figure 1A,B and Figure S8. One of these clusters occupied almost the entire Galicia, while the others had a more localized distribution (figure 3 in Pardo‐Seco et al. [2]).

From a national perspective, Pardo‐Seco et al. [2] analyzed an independent GWAS database (referred to as NDNAB‐PS in [2]) that provides broad representation of Iberian provinces and includes an amount of SNPs comparable to those used by Bycroft et al. [1]; albeit with differences in geographic sampling coverage between the two studies. Rather than detecting the pronounced East–West genetic patterns reported by Bycroft et al. [1] (Figure 2A), Pardo‐Seco et al. [2] identified a main cluster covering the whole South‐Central Iberian Peninsula, spanning East to West across Andalusia, along with well‐localized clusters throughout most of the Spanish territory (Figure 2B; Figure S9). These findings suggest a markedly different pattern of population structure from that reported by Bycroft et al. [1], despite the use of datasets with broadly comparable genomic resolution.

FIGURE 2.

Comparison of the contrasting patterns of genetic diversity reported by Bycroft et al. [1] and Pardo‐Seco et al. [2] for the national Spanish population. (A) Map of clusters reported by Bycroft et al. [1] in their figure 1B, indicating a major East–West genetic divergence across a relatively homogeneous North–South pattern across the Iberian Peninsula. (B) Modified version of the map reported by Pardo‐Seco et al. [2], based on a comparable Spanish dataset (NDNAB‐PS) from the perspective of biogeographical information availability. This map indicates more well‐localized clusters in northern Iberia, consistent with geographic proximity and known demographic history, and a broader cluster spanning Andalusia and large portions of the central Iberian Peninsula. Although the sampling in Pardo‐Seco et al. [2] is more geographically limited than that in Bycroft et al. [1], it is still sufficient to highlight the lack of reproducibility of the reported East–West genetic divergence. (C) Reproduction of figure 1C from Bycroft et al. [1], which illustrates the changes in linguistic and political boundaries in the Iberian Peninsula between approximately 930 and 1300 ce, based on Baldinger [3]. As noted by the authors, “only the colors and labels of the Christian kingdoms have been added to aid visualization.” This panel is included for comparison with the genetic patterns shown in (A) and (B), as discussed in the main text and Supporting Information.

Moreover, consistent with earlier mitochondrial DNA (mtDNA) and the Y‐chromosome studies [10, 11, 12, 13, 14], Pardo‐Seco et al. [2] found that the first hierarchical split separates the Basque Country from the neighboring Cantabria region, while Galicia appears in the final division of the tree. This pattern is in line with prior evidence suggesting that Galicia was likely one of the last regions of continental Europe to be populated in ancient times.

Bycroft et al.'s [1] attribution of the observed East–West divergence and North–South genetic continuity in the Iberian Peninsula to the historical process of the Reconquista [sic] remains highly speculative: “Our observation that genetic differences are small in the north‐south direction within Spain, and evidence of gene flow preferentially in this direction, are most straightforward to interpret in the light of historical information regarding the Reconquista, during with Christian‐controlled territory in the north moved gradually southwards from the mid‐8th Century, following the Muslim conquest of Iberia…” [1]. The hypothesis that the gradual territorial expansion of Christian kingdoms progressed from North to South may help explain a pattern of genetic homogeneity along this axis (provided that “North” is understood as referring to regions South of the big Cantabrian Mountains, rather than the Cantabrian coast itself). However, this is difficult to reconcile with the pronounced East–West genetic differentiation described by the authors and also the fact that most significant geographic barriers (rivers and mountain ranges) in Iberia run along an East–West axis, which would facilitate movements in this direction while slowing down movements from North to South (Supporting Information). Linking these genetic patterns directly to the territorial expansion of Christian kingdoms would require assuming large‐scale population movements from North to South, while simultaneously maintaining marked East–West genetic differentiation at a much finer geographic scale; this assumption is geographically lacks historical support. For instance, one of the largest resettlement initiatives of the Christian territorial expansion, the ninth‐century Repoblación (“repopulation”), involved the settlement of a vast territory between the River Duero and the southern slopes of the Cantabrian Mountains from East to West. Given its scale, this process likely increased regional gene flow, making it difficult to reconcile with the claimed East–West genetic pattern. Furthermore, Bycroft et al. [1] observed that “the broad‐scale genetic structure of the region is strikingly similar to the linguistic frontiers²⁹ present in the Iberian Peninsula around 1300 CE” (see their Figure 1C, based on maps by Baldinger [3]; reproduced here as Figure 2C). Interestingly, this historical linguistic division appears to more closely resemble the genetic pattern described by Parso‐Seco et al. [2], in which a major lineage is almost uniformly distributed across much of the central and southern Iberian Peninsula, than to the pattern reported by Bycroft et al. [1], further complicating any straightforward interpretation of the genetic landscape in terms of this period of Christian expansion alone.

Finally, Pardo‐Seco et al. [2], along with recent ancient DNA studies [15, 16], and well‐established historical and archeological evidence (see [2] for an in‐depth discussion), bring together multiple lines of evidence that challenge the traditional hypothesis attributing North African and Middle Eastern ancestry in the northern Iberian Peninsula primarily or solely to this period of Christian expansion. Instead, these findings suggest the need to reconsider the role of earlier, pre‐Islamic contacts, dating back at least to Roman times, as significant contributors to the demographic landscape that shaped the genetic structure of northern Iberian populations, particularly in Galicia. This perspective represents a paradigm shift in understanding the historical processes underlying present‐day genomic patterns in the region.

2.4. FineSTRUCTURE Might Be Highly Sensitive to Genetic Stratification, Whether Real or Artificial

Bycroft et al. [1] used fineSTRUCTURE software to detect fine‐scale genetic patterns, a method that is more sensitive than traditional approaches, such as principal component analysis (PCA). While this sensitivity enables the identification of subtle demographic processes, it may also increase susceptibility to technical artifacts introduced during laboratory processing, even when such artifacts are not apparent using standard methods. To illustrate this, we conducted a few pragmatic in silico experiments (see Supporting Information for further details).

Before addressing this issue, we first test the sensitivity of fineSTRUCTURE to sample size. In their analysis of the entire GAL cohort, Pardo‐Seco et al. [2] identified five clusters containing at least four individuals: ‘Main’, ‘Porto do Son’, ‘Cuntis/A Estrada’, ‘A Mariña/Foz’, and ‘Central/West’. To assess the effect of sample size, we ran fineSTRUCTURE on the GAL₃ dataset, which consisted of a random selection of only three individuals from each of these five clusters. Despite the substantial reduction in sample size, fineSTRUCTURE largely reproduced the topology of the original clustering tree, suffering only one minimal modification. Specifically, the analysis retained three of the main clusters and merged the ‘Main’ and ‘A Estrada/Cuntis’ clusters into a single group (Figure 3A). These results suggest that reducing sample size does not necessarily hinder cluster detection or substantially alter tree topology in our case, as long as a small but representative sample from each cluster is included.

FIGURE 3.

Comparison of dendrograms and PCA analyses derived from the GALOMICS (GAL) dataset and the NDNAB‐PS dataset, illustrating the ability of fineSTRUCTURE to detect fine‐scale population structure under reduced sample sizes and its sensitivity to batch effects. (A) Binary tree of GAL₃ (a reduced dataset with three random selected individuals from each GAL cluster), used to test the effect of sample size on fineSTRUCTURE performance, showing overall topological coherence with the clustering pattern obtained from the full GAL dataset (Figure 1B); (B) fineSTRUCTURE analysis of the NDNAB‐PS dataset, highlighting the position of Galician samples at the tip of the tree and effectively localizing the three main regions of the Cantabrian coast, with clustering coherence consistent with historical and geographical patterns (a heatmap of pairwise Total Variation Distance (TVD) for the NDNAB‐PS dataset is provided in Figure S9). For panels (B), the terminal (collapsed) branches are labeled by the dominant geographic region contributing to the clade (e.g., Galicia, Andalucia, Catalonia), with percentages indicating the proportion of individuals from that region. In square brackets, the number of terminal branches and their sample sizes are reported. For example, the top blue branch is predominantly Galician (96.4%), comprising 56 individuals distributed across three terminal branches of sizes 46, 7, and 3; blue coloring (matching the region name, Galicia) indicates that the clusters of sizes 7 and 3 consist exclusively of Galician individuals, while black coloring is used to indicate mixed branches (the terminal branch of size 46 includes individuals from Galicia and other Iberian regions). The PCA plots in panels (A)–(B) were generated from the same datasets used for the corresponding dendrogram analyses and compare a conventional PCA based on the variance‐standarized relationship matrix with a PCA derived from the ChromoPainter haplotype‐sharing matrix. The inferred fineSTRUCTURE clusters align with the primary geographic structure of Iberia and the major dendrogram clades.

To assess the sensitivity of fineSTRUCTURE to batch effects, we analyzed a merged dataset combining samples from the GAL (WGS) and NDNAB‐PS (genotyping array) cohorts, which were generated using different technological platforms and processed in independent laboratories. In the absence of significant batch effects, GAL samples cluster with Galician individuals from the NDNAB‐PS dataset, reflecting shared geographic origin. In contrast, fineSTRUCTURE produced a markedly different and more complex structure. Before merging, Galician individuals in NDNAB‐PS formed a distinct peripheral cluster in the fineSTRUCTURE dendrogram (Figure 3B). In this tree, Andalusia appears as the most closely related cluster, an unexpected result given the geographic proximity of Asturias and its likely more similar demographic history.

After integrating the NDNAB‐PS and GAL datasets, all Galician samples from both cohorts shifted to the deepest internal node in the dendrogram (Figure 4A), becoming separated from the remaining Iberian samples. This shift suggests an artificial signal of genetic differentiation that is difficult to reconcile with current demographic, historical, and population‐genetic knowledge of the region. Consequently, it is more plausibly explained by technical artifacts arising from the integration of datasets generated using different technologies and analytical pipelines. This mirrors the pattern reported by Bycroft et al. [1], where unexpected deep splits are likely reflecting technical artifacts rather than true population structure.

FIGURE 4.

Comparison of PCA analyses and dendrograms derived from the merged GALOMICS (GAL) and NDNAB‐PS datasets, illustrating the sensitivity of fineSTRUCTURE to batch effects and their impact on inferred population structure. (A) Distortion of clustering after integration of GAL and NDNAB‐PS samples, with Galicians shifting to a deeper internal node in the tree, separating the profiles primarily into two batches (GAL and NDNAB‐PS), which is inconsistent with demographic and geographical expectations (an individual‐sample extended dendrogram is provided in Figure S10). (B) Incoherent clustering generated after combining GAL and Galicians from NDNAB‐PS, with samples separated by processing batches rather than by biogeographical/genetic proximity (extended dendrogram in Figure S11). The figure additionally indicates the proportion of samples contributed by the NDNAB‐PS and GAL datasets; for instance, in panel (A), the three bottom clades are mainly composed of Galicians, with one clade composed of 100% GAL samples, another of 90.4% NDNAB‐PS samples, and a third of 89.4% GAL samples. The PCA plots in panels (A)–(B) were generated from the same datasets used for the corresponding dendrogram analyses and compare a conventional PCA based on the variance‐standarized relationship matrix distances with a PCA derived from the ChromoPainter haplotype‐sharing matrix. See the legend of Figure 3 for additional information common to the panels.

Most importantly, the merged Galician cluster itself warrants closer scrutiny. Specifically, it is divided into two major subclusters and one minor subcluster. One major subcluster and the minor one were composed almost entirely of GAL profiles (~92% and ~100%, respectively), while the other major subcluster consisted predominantly of NDNAB‐PS samples (~90%) (Figure 4A). Notably, each subcluster included individuals from multiple Galician provinces, indicating a lack of geographical coherence. This pattern strongly suggests that the observed structure is driven by technical batch effects rather than true fine‐scale geographic differentiation.

In addition, the merged dataset (based on a common set of > 599K SNPs) yielded significantly more clusters among Galicians (n = 17 clusters; Figure S10) than those identified in the individual fineSTRUCTURE analyses of the GAL dataset using 17.2 million variants from WGS (n = 10 clusters, 5 of which included at least 4 individuals; Figure 1B) and the NDNAB‐PS dataset using > 599K SNPs (n = 3 clusters). This suggests that merging datasets processed through different pipelines may lead to an artificial inflation of inferred population structure.

Finally, when we combined GAL (representing exclusively Galician samples) with only the Galicians in NDNAB‐PS, again, fineSTRUCTURE increasingly differentiated the samples based on batch rather than genetic characteristics, further emphasizing the software's sensitivity to batch effects (Figure 4B, Figure S11).

For the different datasets and combinations, we additionally explored the differences between conventional PCA based on the variance‐standarized relationship matrix and a PCA derived from the ChromoPainter haplotype‐sharing matrix (Figure 3A,B and Figure 4A,B). In each case, the PCA analyses were performed using the same genetic and sampling data used to construct the corresponding dendrograms. As expected, conventional PCA showed reduced sensitivity for resolving fine‐scale population structure, resulting in greater overlap among samples in Euclidean space. Nevertheless, it was still able to detect evidence of batch effects between the GAL and NDNAB‐PS datasets (e.g., Figure 4B). By comparison, PCA based on the haplotype‐sharing matrix provided a clearer separation of genetic clusters and more closely recapitulated the patterns observed in the corresponding fineSTRUCTURE dendrograms, reflecting the greater power of haplotype‐based approaches to resolve subtle population structure.

3. Discussion

A meticulous investigation of the findings reported by Bycroft et al. [1] revealed multiple lines of evidence suggesting that technical and methodological issues may have contributed to the unexpectedly large number of genetic clusters identified in Galicia. While the authors had previously acknowledged batch effects in earlier publications, this issue was not explicitly addressed in their 2019 study. We also identified anomalous patterns of genetic divergence at the national level that may reflect the same technical influences.

To further investigate this hypothesis, we conducted additional analyses using independent datasets. These analyses show that, although fineSTRUCTURE is a powerful method for detecting fine‐scale population structure, it may be highly sensitive to technical variation that may remain undetected by standard quality control approaches such as PCA. As a haplotype‐based approach, fineSTRUCTURE relies on coancestry matrices that can be vulnerable to systematic biases introduced during variant calling, batch‐specific processes (e.g., DNA extraction, library preparation, sequencing runs), and imputation errors, all of which can lead to artificial clustering.

Further research is needed to understand how this sensitivity to batch effects may influence not only population structure inference but also downstream applications such as admixture dating, as performed by tools like GLOBETROTTER. For example, while Bycroft et al. [1] attribute the main North African genetic contribution in Iberia to the Reconquista [sic], recent evidence from ancient DNA presented by Oteo‐García et al. [15] points to a different scenario. Their analysis of medieval samples from Valencia suggests that the North African component was already established before the Christian reconquest, indicating genetic continuity rather than a post‐Reconquista [sic] influx. Recently, Roca‐Rada et al. [17] demonstrated “that the village of Idanha‐a‐Velha (Portugal) emerges in the Roman period as a site of significant migration and interaction, presenting a notably diverse genetic profile that includes North African and Eastern Mediterranean ancestries”. Several Roman‐period individuals from this village show genetic shifts toward Africa in PCA and f‐statistics; their results suggest that North African ancestry was present in this village during the Roman period, long before Islamic rule. Similar conclusions were reached by Pérez‐Ramallo et al. [16] on the analysis of the bone remains of the Galician bishop Teodomiro. Additionally, findings by Pardo‐Seco et al. [2] suggest that the main episode of North African and Middle Eastern genetic influence in Galicia and the broader northern Iberian Peninsula may have occurred well before the Islamic‐ruled territories and Christian expansion, supporting a more complex and regionally sophisticated demographic history. There are other recent pieces of evidence in ancient DNA studies pointing to North African ancestry in Iberia before the Islamic rule [18, 19].

Apart from the sensitivity of fineSTRUCTURE to technical artifacts, we have also identified additional issues in the study by [1], including potential sampling bias within the EPICOLON cohort. Specifically, some regions appear to be disproportionately represented by individuals from rural areas (e.g., Galicia, Basque Country), whereas others are predominantly represented by urban populations (e.g., Catalonia, Andalusia). Such heterogeneity in sampling strategies may introduce interpretative biases and influence the clustering patterns observed across Spanish regions, potentially complicating the distinction between genuine population structure and differences arising from sample composition.

Little can be added beyond what has already been extensively discussed in the literature regarding the identification and mitigation of batch effects. Standard exploratory approaches for batch‐effect detection, such as PCA, are often sufficient to reveal systematic technical structure and to motivate appropriate corrective actions, particularly when batch effects are relatively strong. A major concern in [1] is that these standard quality‐control analyses were either not reported, discussed, or adequately incorporated into the interpretation of the findings. Importantly, evidence of batch effects was already documented by the authors in earlier publications, where conventional exploratory analyses revealed technical stratification (Supporting Information). Thus, the central issue in [1] is not whether batch effects are present in these datasets, but rather whether their potential impact was considered during data processing and interpretation of the observed population structure.

The broader challenge for the field is to develop and apply methodologies capable of identifying more subtle forms of technical variation that may not be readily apparent in standard exploratory analyses, yet can become problematic when using highly sensitive clustering approaches designed to detect very fine‐scale population structure, such as fineSTRUCTURE. Methods of this type are specifically intended to identify subtle genetic subdivisions and, consequently, may also be particularly sensitive to low‐level technical artifacts. For this reason, careful assessment of residual batch effects becomes increasingly important as the resolution and sensitivity of population‐genetic analyses continue to increase.

In summary, Bycroft et al. [1] claimed to identify ultra‐fine‐scale population structure in Galicia and across Spain, but their investigations suffer from unaccounted technical biases (explained in greater detail in the Supporting Information). The extreme genetic stratification reported in Pontevedra contradicts historical, geographical, and demographic evidence and appears highly counterintuitive, especially considering the historical and present‐day cosmopolitan nature of the region, as well as the absence of geographical barriers (contra [1]). At the national level, unexpected patterns of clustering and genetic differentiation across Iberia, together with the failure of independent high‐resolution genomic datasets to reproduce these findings, suggest that hospital‐specific batch effects may have distorted their findings. Inconsistencies in sampling methods further raise concerns about the reliability of their conclusions. Moreover, because several subsequent studies relied on the same EPICOLON dataset, potential technical artifacts may have propagated into later interpretations of Iberian population structure. Importantly, the purpose of the present work is not to categorically dismiss the possibility that some degree of fine‐scale population structure exists in Galicia or elsewhere in Iberia. Rather, we argue that such claims require robust empirical support and independent validation. At present, the available evidence from independent datasets appears inconsistent with the magnitude of structure reported by [1] and, more generally, contrasts with the prevailing conclusions of previous population‐genetic studies conducted in the region.

Future genetic studies should prioritize methodological transparency, control for potential technical artifacts, and implement strategies to detect and mitigate batch effects when using highly sensitive methods like fineSTRUCTURE to ensure more accurate representations of human population structure.

4. Materials and Methods

4.1. Genome Datasets

We used the genomic data as in Pardo‐Seco et al. [2]. The Whole‐Genome Sequencing (WGS) dataset GALOMICS (GAL; n = 91) represents a primarily rural cohort, where all participants have deep local ancestry. The average distance between each donor's birthplace and that of their four grandparents is 6.6 km (SD: 15.8), with a median of 0 km (IQR: 0–5.5 km). Similarly, the average distance among the birthplaces of the four grandparents is 5.4 km (SD: 10.7), with a median of 0 km (IQR: 0–5.2 km). We also used Iberian samples from the National DNA Data Bank (n = 470 donors), as done by Bycroft et al. [1]. These samples are hereafter referred to as NDNAB‐PS (to distinguish them from the NDNAB dataset in that study), and are characterized by geographic resolution at the provincial level (a ‘province’ in Spain is roughly equivalent to a ‘county’ in both area and population). Although not explicitly stated by the authors, this same level of resolution is suspected for most samples in Bycroft et al. [1]; Supporting Information. Also, according to the metadata provided by the Spanish National DNA Bank, it is not possible to determine whether the location refers to the name of the province or to its capital city; the impact of this feature most likely extends to the EPICOLON dataset (Supporting Information).

Methodological details not described here, including the generation and processing of the genomic datasets, quality‐control (QC) procedures, filtering criteria, variant calling, imputation, and sample‐selection procedures for both the GAL and NDNAB‐PS datasets, are provided in detail in Pardo‐Seco et al. [2]. The present study uses the same processed datasets and follows the methodological framework described therein unless otherwise stated.

4.2. fineSTRUCTURE Analyses

Clusters were inferred using fineSTRUCTURE following recommendations provided by the authors [20]. The number of clusters is estimated based on the underlying probability model. To estimate the coancestry matrix derived from the clusters and construct a hierarchical clustering dendrogram of the clusters based on haplotype‐sharing similarity, we used the CHROMOPAINTER package [20]. FineSTRUCTURE was run on both the GAL and NDNAB‐PS datasets to investigate population clusters within Galicia and at a national scale.

The results were compared with those of Bycroft et al. [1]; these analyses aim to highlight the severe contrast between the low‐level population structure observed using the GAL and NDNAB‐PS datasets and the extraordinarily fine‐scale population structure reported by [1], providing further insight into these discrepancies.

The Total Variation Distance (TVD) matrix was computed as half of the sum of absolute differences between the coancestry probability distributions for each pair of individuals, using coancestry chunk‐length files generated with ChromoPainter. These values were visualized as heatmaps (Figure S8, Figure S9), considering the clustering structure inferred with fineSTRUCTURE for both GAL and NDNAB‐PS.

We carried out two experiments using fineSTRUCTURE to test its sensitivity to specific characteristics of the data:

1. Sensitivity to sample size: We ran fineSTRUCTURE on a subset of the GAL dataset, randomly selecting only three samples from each of the five clusters identified in the full GAL dataset (referred to in the text as GAL₃) as reported by Pardo‐Seco et al. [2]. The goal was to determine whether sample size affects the number of clusters detected, assuming the clusters exist. While the GAL sample size is smaller than that of EPICOLON, we hypothesize it is still sufficiently large to detect most, if not all, of the clusters present in the population, particularly given that we are using WGS data, as opposed to the lower‐resolution panel of ~693 K SNPs used by Bycroft et al. [1].

2. Sensitivity to batch effects: In the present study, we hypothesize that the anomalous results reported by Bycroft et al. [1] may have been strongly influenced by batch effects, compounded by the high sensitivity of fineSTRUCTURE in detecting fine‐scale structure in genomic datasets, whether real or artefactual. To investigate this, we applied fineSTRUCTURE to a merged dataset composed of the GAL and NDNAB‐PS datasets, which were generated using two different technical procedures (GAL: next‐generation sequencing [NGS]; NDNAB‐PS: genotyping). Two different runs were carried out: (i) merging GAL + NDNAB‐PS and (ii) merging GAL (all Galicians) with only the Galicians present in the NDNAB‐PS dataset. These two simple exercises allowed us to evaluate the behavior of fineSTRUCTURE under deliberately induced conditions of batch stratification.

4.3. Visualization of Genetic Data in Maps

Individual frequencies for the main genetic clusters identified in the NDNAB‐PS dataset were calculated across the various sampling points. These frequencies were then spatially represented on a geographic map using SAGA GIS v. 9.6.1 (http://www.saga‐gis.org/) [21] and interpolated using the ordinary Kriging method. Frequencies for each genetic cluster were interpolated separately and subsequently combined into a single composite layer to provide a clearer and more integrated overview of cluster distribution. Each data point on our map (Figure 2B) represents multiple individual samples (sharing the same coordinates). Given that our analysis is conducted at the provincial level, points are centered on provincial capitals. This contrasts with the approach taken by Bycroft et al. [1], who applied a random spatial shift of 24 km to samples sharing the same coordinates to facilitate visualization. This method introduces an artificial sense of spatial sampling and geographic dispersion, potentially obscuring the underlying biogeographic patterns present in the EPICOLON dataset (Supporting Information).

Author Contributions

A.S. and F.M.‐T. conceived the project and contributed the funding. A.S., J.P.‐S., A.G.‐C., and C.M.‐C. carried out the analysis. A.S. wrote the draft of the study. All the authors critically reviewed the final version of the manuscript.

Funding

This study also received support by (i) ISCIII: TRINEO: PI22/00162; DIAVIR: DTS19/00049; Resvi‐Omics: PI19/01039 (to A.S.), ReSVinext: PI16/01569, Enterogen: PI19/01090, OMI‐COVI‐VAC: PI22/00406 (to F.M.‐T.), cofinanciados FEDER, (ii) GAIN: IN607B 2020/08 and IN607A 2023/02 (to A.S.), GEN‐COVID IN845D 2020/23 (to F.M.‐T.), IIN607A2021/05 (to F.M.‐T.) and IN677D 2024/06 (to A.G.‐C.); (iii) ACIS: BI‐BACVIR (PRIS‐3, to A.S.), CovidPhy (SA 304 C, to A.S.), PneumoTrack (PRIST‐VAL, to A.S.), and Respisal (PRIST‐VAL, to F.M.‐T.); (iv) consorcio Centro de Investigación Biomédica en Red de Enfermedades Respiratorias (CB21/06/00103; to A.S. and F.M.‐T.) and (v) Spanish Ministry of Science and Innovation (MCIN)/Spanish Research Agency (AEI) (KAWA‐TesT: PID2022‐142156OB‐I00, to AG‐C). This study has been funded by Instituto de Salud Carlos III (ISCIII) through the project “CP23/00080” and co‐funded by the European Union (AG‐C). This work was supported by the European Union's Horizon 2020 research and innovation programme under Grant Agreement Nos. 668303 (PERFORM) and 848196 (DIAMONDS). The funders were not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Anomalies in the Galician bin ary tree and fine‐level genetic stratification. (A) Full tree according to Figure 1A in Bycroft et al. (1); it is quite straightforward to count the branches in non‐Galician regions, which amount to approximately 38. This suggests that in Galicia, there are 145 minus this quantity (107 branches), representing > 73% of the total Iberian clusters concentrated in Galicia, primarily in the small region of southwestern Pontevedra province. (B) A capture of the map from Figure 1 of Bycroft et al. (1), highlighting the Galician sampling area with the yellow border indicating the area corresponding to the province of Pontevedra. All samples were collected at Hospital do Meixoeiro, located in a well‐connected region surrounding the port city of Vigo. The biogeographical origins of most samples depicted on the map are concentrated within a 60 km area East of Vigo and 20 km South of this cosmopolitan region, with no significant geographical barriers in the vicinity. A zoomed‐in view of the rectangular area in (A), extracted from their Figure 3A, allows for a direct comparison. The top portion of the figure (C) presents a magnified section designed to facilitate an exact regional comparison between Figures 1 and Figure 3A from Bycroft et al. (1). We have edited their figures to more clearly illustrate the sampling points shown in their Figure 3A (the zoomed‐in version), which provides finer‐scale geographic detail compared to the more diffuse clustering pattern observed in their Figure 1. The white lines in (B) indicate a well‐connected corridor around Vigo, along with the distances. For reference, we indicate a circle with a radius of 24 km centered on Hospital do Meixoeiro (on the left), which is presumably used to randomly disperse samples sharing the same coordinate. Additionally, we show the approximate 3 km diameter (zoomed‐in map on the right) that encompasses the actual cluster located at Hospital do Meixoeiro. Key localities along this axis (highlighted with red and yellow circles) and their estimated population sizes (according to official census https://www.ige.gal) are Vigo (~293.6K), Tomiño (~13.9K inhabitants), Tui (~17.5K), Porriño (~20.7K), A Guarda (~10.1K), and Ponteareas (~23.2K). Therefore, these are not small villages but rather urbanized areas (small cities) with strong transport links to Vigo and the nearby Portuguese border. (D) A capture of their Figure 3A map where we highlight the presence of a single member of the Galician cluster > 500 km away (in the Basque Country) from its apparent genuine location in the Pontevedra locality of A Guarda. Additional notable examples of ‘delocalized’ genuine clusters are provided in Figure S7.

Figure S2: Full binary tree from Figure 1A (top) and a section of the map from Figure 1B (bottom) in Bycroft et al. (1), highlighting the unusual pattern that emerges when comparing the high genetic complexity observed in the cosmopolitan area of Vigo (Pontevedra) with the marked genetic homogeneity found on both sides of the Cantabrian Mountains. The representation also highlights notable clustering inconsistencies in the clusters described for the northern Cantabrian regions, where geographically adjacent areas (e.g., Asturias and Cantabria, separated by ~160 km) appear genetically distant in the inferred structure, while showing unexpected genetic similarity to geographically distant regions (e.g., Catalonia and the Balearic Islands). Arrows and outlined areas indicate the genetic clusters discussed in the text. Also included is a schematic representation of the Cantabrian Mountains and the Basque Mountains, which separate the northern Iberian coast from the rest of the Peninsula.

Figure S3: Contrasting patterns between the North‐South genetic continuity and East–West divergence patterns in the Iberian Peninsula as described by Bycroft et al. (1), in relation to major East–West geographic barriers. (A) Modified version of Figure 2A from Bycroft et al. (1), showing the distribution of major genetic clusters across the Iberian Peninsula from North to South, overlaid with a schematic representation of key geographic barriers (blue lines: main rivers and valleys; white lines: large mountain ranges). (B) Topographic map of the Iberian Peninsula, highlighting the most prominent natural barriers, including major mountain ranges and river systems, which predominantly extend in an East–West direction. Source for map in panel (B): https://commons.wikimedia.org/wiki/File:Spain_topo.jpg.

Figure S4: Comparative clustering patterns in (A) García‐Calleja et al. (16) (see also their Table S1), (B) Pardo‐Seco et al. (2) (present study, Figure 2B in the main text), and (C) Bycroft et al. (1) (modified capture from their Figure 2A), highlighting differing interpretations of East–West genetic patterns in García‐Calleja et al. (16) and Bycroft et al. (1). In panel (A), we also include a capture of the Iberian clustering tree from García‐Calleja et al. (16), as shown in their Figure S6. García‐Calleja et al. (16) noted that samples from Valencia appear in both the East and West branches, although they did not specify how many fall into each group. The individuals from Eivissa (Ibiza) were sampled through their ‘Cat array’ dataset; however, the regional classification of other Balearic Island samples within this dataset, specifically from Mallorca (n = 9) and Menorca (n = 3), was not discussed. Other samples from the ‘Cat array’ dataset also lack clustering assignment. Paradoxically and counterintuitively, the Eivissa samples are assigned to the West branch, which further undermines the plausibility of the proposed East–West genetic gradient comparable to the one described Bycroft et al. (1). Most individuals represented in the map (A) originated from the GCAT cohort (n = 141), with the majority coming from Catalonia in the northeastern Iberian Peninsula and Andalusia, which spans the entire southern region of the Peninsula from East to West. Additional notes: García‐Calleja et al. (16) did not specify the East/West branch assignment for samples from Galicia (n = 4), Madrid (n = 1), and Murcia (n = 3). Furthermore, several northern autonomous communities (Asturias, Cantabria, Navarra, La Rioja, and the Basque Country) were not represented in their dataset.

Figure S5: Edited map and binary tree adapted from Figure 2A of Bycroft et al. (1), in which Iberian samples (EPICOLON) are analyzed alongside a Portuguese sample. The highlighted clustering pattern reveals a distorted structure introduced by the inclusion of the Portuguese sample, resulting in an artificial division of Galician samples into two major, non‐sister branches. One branch, representing the deepest split in the tree, includes most of the minor Galician clusters; > 105 as inferred from Bycroft et al. (1). The second Galician/Portuguese branch is positioned next at an internal node shared with multiple Iberian clusters, including both geographically adjacent regions and more distant ones, such as the Balearic Islands. Horizontal red dotted lines mark the first, second, and third splits in the tree, emphasizing the considerable genetic distance between the two main Galician branches.

Figure S6: Maps illustrating that the EPICOLON cohort used by Bycroft et al. (1) was largely based on urban sampling, primarily from some of the most cosmopolitan cities in Spain. (A) Map of the Iberian Peninsula showing Spain's first‐level political and administrative divisions, including 17 autonomous communities, two autonomous cities, and the locations, boundaries, and capitals of the 50 provinces (source: https://www.nationsonline.org/oneworld/map/spain‐administrative‐map.htm). (B) A modified version of Figure 1 from Bycroft et al. (1), highlighting that the main clusters of sampled individuals are concentrated around the capital cities of Spanish provinces and other big cities. This pattern suggests that sampling in the study was conducted primarily at a province‐level geographic resolution, with most samples likely representing urban populations. Red circles indicate the primary recruitment areas (eleven hospitals) according to Fernández‐Rozadilla et al. (3); in most locations, a single hospital was involved, except for Barcelona, which had four recruiting hospitals. Blue circles indicate additional recruiting hospitals reported by Castellví‐Bel et al. (4). Black circles mark other sample clusters concentrated around urban areas for which no specific hospital has been identified. The increased density of black circles overlapping the yellow band (representing the ‘yellow’ clusters in the map) is consistent with our hypothesis that many of these samples were likely obtained through the Spanish National DNA Bank (NDNAB). The circles have an approximate radius of 24 km, which corresponds to the distance used by the authors to randomly displace samples sharing the same coordinates.

Figure S7: Modified Figure 1 from Bycroft et al. (1) aimed at reconstructing batch patterns in this study. (A) Circles represent scattered occurrences across various areas of Spain for clusters that appear to be well localized in specific regions. This pattern suggests that, rather than reflecting genuine genetic structure, cluster assignment may have been influenced by technical batch effects. For instance, blue circles represent ‘genuine’ Basque Country clades that appear within the North‐South, red‐dominated band in the figure, suggesting that at least these samples were processed as a batch at Hospital de Donostia in the Basque Country; (B) Overlay of the main recruiting hospitals for the EPICOLON samples, with various colored shapes indicating our inference of the primary batch effects associated with each hospital. Clusters are named as in Bycroft et al. (1).

Figure S8: Heatmap of pairwise Total Variation Distance (TVD) for the GAL dataset, based on the fineSTRUCTURE clustering reported by Pardo‐Seco et al. [5].

Figure S9: Heatmap of pairwise Total Variation Distance (TVD) for the NDNAB‐PS dataset, using the fineSTRUCTURE clustering dendrogram shown in Figure 3B.

Figure S10: Extended fineSTRUCTURE individual‐level dendrogram obtained from the joint analysis of GAL samples and Galician individuals from the NDNAB‐PS dataset; the corresponding clustered dendrogram is shown in Figure 4A.

Figure S11: fineSTRUCTURE dendrogram showing the individual‐level distribution of Galician samples from the GAL and NDNAB‐PS datasets analyzed together. Non‐Galician individuals are collapsed into major clusters, as displayed in Figure 4B.

Appendix A.

GALOMICS Working Group—GenPoB & GenViP IDIS research groups. Principal Investigators: Antonio Salas Ellacuriaga, and Federico Martinon‐Torres.

Alba Camino Mera, Alberto Gómez Carballa, Alicia Carballal Fernández, Ana Cotovad Bellas, Ana Isabel Dacosta Urbieta, Ana María Pastoriza Mourelle, Ana María Senín Ferreiro, Andrés Efraín Muy Pérez, Antía Rivas Oural, Antonio Justicia, Grande, Antonio Piñeiro García, Antonio Salas Ellacuriaga, Anxela Cristina Delgado García, Ángela Manzanares Casteleiro, Belén Mosquera Pérez, Blanca Díaz Esteban, Carlos Durán Suárez, Carmen Curros Novo, Carmen Rodríguez‐Tenreiro Sánchez, Conrado Martínez Cadenas, Cristina Serén Trasorras, Cristina Talavero González, Esther Montero Campos, Federico Martinón‐Torres, Fernando Álvez González, Fernando Caamaño Viña, Gloria Viz Rodríguez, Hugo Alberto Tovar Velasco, Irene Rivero Calle, Irene Visos Varela, Isabel Ferreirós Vidal, Isabel Rego Lijo, Iván Prieto Gómez, Jacobo Pardo Seco, Jesús Eirís Puñal, José Gómez Rial, José Manuel Fernández García, José María Martinón Sánchez, Julia García Currás, Julián Montoto Louzao, Lara Martínez Martínez, Laura Navarro Marrón, Lidia Piñeiro Rodríguez, Lorenzo Redondo Collazo, Lúa Castelo Martínez, Manuel Vázquez Donsión, María Elena Gamborino Caramés, María Elena Sobrino Fernández, María José Currás Tuala, María Martínez Castro, María Soledad Vilas Iglesias, María Sol Rodriguez Calvo, María Sonia Ares Gómez, María Teresa Autrán García, Marina Casas Pérez, Marta Aldonza Torres, Marta Bouzón Alejandro, Marta Lendoiro Fuentes, Miguel Sadiki Orayyou, Miriam Ben García, Miriam Cebey López, Miriam Taboada Puga, Monserrat López Franco, Narmeen Mallah, Natalia García Sánchez, Nour El Zahraa Mallah, Ouhao Zhu Huang, Regueiro Casuso, Rosaura Picáns Leis, Ruth Barral Arca, Sandra Carnota Antonio, Sandra Viz Lasheras, Sara Pischedda, Sara Rey Vázquez, Silvia Cerqueiro Rey, Sonia Marcos Alonso, Sonia Serén Fernández, Susana María Rey García, Ulises Toscanini, Vanesa Álvarez Iglesias, Wiktor Dominik Nowak, Xabier Bello Paderne.

Data Availability Statement

The analyses described in the present study were conducted without the generation of new data.