Signals of positive selection outside Africa

We first explored the genomic signatures of the selective sweep previously reported in the SLC30A9 gene region [14,16–18,20,25]. Hence, we compiled evidence from different summary statistics and selection tests available at the PopHuman browser for the 26 human populations from the 1000 Genomes Project (1000 GP; http://pophuman.uab.cat) [26]. To summarize and visualize the continental patterns of positive selection along the SLC30A9 region, we extracted the F_ST, XP-EHH, and Fay and Wu’s H values in three populations used as a reference for Africa (Yoruba: YRI), Europe (Residents of Utah with North and Western European ancestry: CEU), and Asia (Han Chinese: CHB) (Fig 1A). Unusual high levels of population differentiation were confirmed along the SLC30A9 region when comparing CEU and CHB with YRI. Similarly, when applying the XP-EHH statistic, we detected genome-wide departures for long haplotypes in CEU and above all in CHB (using YRI as the reference). Furthermore, a significant excess of derived alleles was detected around the SLC30A9 region in both CHB and CEU populations when analyzing the site frequency spectrum with the Fay and Wu’s H statistic. Notably, most of these signatures of selection replicate for the remaining 1000 GP populations available at the PopHuman browser in all continents except Africa.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Signals of positive selection at SLC30A9.

(A) Signatures of adaptation in the SLC30A9 region. Chromosomal location and genes are displayed at the top. In the first selection panel, window-based F_ST values when comparing CEU versus YRI and CHB versus YRI are represented by cyan and pink bars, respectively. Similarly, window-based Fay and Wu’s H values and XP-EHH scores (using YRI as the reference population) in CEU and CHB are shown with the same color scheme in the middle and bottom selection panels, respectively. Overlapping population scores can result in blue shades. Horizontal cyan and pink lines represent the genomic average of each test for each population, whereas 2 standard deviations are represented by the background shade in each case as a mean to interpret the genome-wide significance of each selection signature. All selection test values were extracted from the corresponding 1000 GP populations as available at the PopHuman browser (http://pophuman.uab.cat) [26] and represented using the Gviz package [58]. (B) Allele frequency trajectories and selection coefficients inferred with CLUES for rs1047626 in CHB and CEU and for rs4861157 in YRI.

https://doi.org/10.1371/journal.pgen.1010950.g001

We next investigated which variants along the SLC30A9 gene could underlie the detected signals of selection (see complete list in S1 Table). From a total of 354 biallelic SNPs with MAF ≥ 0.02 in SLC30A9 gene, we initially selected as candidates those variants with high population differentiation when comparing YRI with CEU and CHB. From the resulting 235 variants, we further considered as putatively adaptive those that were either non-synonymous, present at the UTR regions of the SLC30A9 gene, or had a CADD Phred score ≥ 10, as such a value predicts that the variant falls within the top 10% of the most deleterious (i.e., functional) variants of the genome [27]. A total of 18 SNPs matched our filtering criteria (Table 1). Among these, we identified a non-synonymous SNP (rs1047626, resulting in the Met50Val substitution), two synonymous variants, five variants at the 3’ UTR, and ten intronic variants. According to the GTEX portal (https://www.gtexportal.org/home/; last accessed 13/04/2022), all 18 putative adaptive variants are eQTLs, with the most common allele in CHB and CEU associated with reduced SLC30A9 expression mostly in brain (S1–S19 Figs). Moreover, iSAFE [28] similarly pinpointed several eQTLs (including rs1047626) among the most likely candidate mutations favored by selection along the region (chr4: 41,853,671–42,153,671; GRCh37) in both CEU and CHB (Tables 1, S2, and S3 and S20 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Putative adaptive variants in the SLC30A9 region.

https://doi.org/10.1371/journal.pgen.1010950.t001

We then took advantage of the previously inferred genome-wide genealogies available for the 1000 GP individuals and used Relate [29] to explore whether the lineages carrying the derived allele of the Met50Val substitution had spread at an unusually fast rate in comparison with competing lineages in the YRI, CHB and CEU populations. The marginal tree corresponding to the surrounding region of rs1047626 displayed the Val-derived variant in multiple lineages concentrated in CHB and CEU, although it was also present in some YRI haplotypes (S21 Fig). In contrast, most of the lineages carrying the ancestral variant were mainly found in YRI. The subsequent analysis with CLUEs estimated a drastic allele frequency shift compatible with the action of positive selection acting on rs1047626 at ~56–77 kya (i.e., 2,000–2,750 generations before the present; using 28 years per generation) with selection coefficients s = 0.0113 (logLR = 5.68) and s = 0.0017 (logLR = 10.93) in CHB and CEU, respectively (Fig 1B). When analyzing other non-African populations such as FIN (European), JPT (East Asian), PJL (South Asian) and PEL (American), we obtained similar allelic trajectories and selection coefficients (S22 Fig). Thus, we seem to be detecting a strong signal of positive selection common to all non-African populations probably dating back before the expansion of modern humans though Europe and Asia.

Potential signals of archaic introgression

We next explored whether such an early selection event could be explained by adaptive introgression. For that, we first investigated whether the Met50Val substitution was present in publicly available Neanderthal and Denisovan genomes. Remarkably, the Denisovan genome presents the homozygous genotype for the derived allele at the rs1047626 position, whereas three high coverage Neanderthal genomes (Altai, Vindija, and Chagyrskaya) are homozygotes for the ancestral allele. We then analyzed the corresponding Neanderthal and Denisovan base positions for a total of 170 SNPs found in high LD (r²>0.8) with rs1047626 in CHB and CEU and spanning a 98.74 kb region (S4 Table). For 164 of these SNPs (96%), the most frequent allele in EAS, EUR, SAS, and AMR differs from the most prevalent in AFR, confirming the existence of two major contrasting Yin/Yang human haplotypes (i.e., two haplotypes differing at each polymorphic site; [30]). On the corresponding 98 covered SNP positions, the Neanderthal and Denisovan genomes were found to display higher allelic similarity and greater sharing of derived alleles with the allelic combinations most frequently found in non-Africans (S23 Fig and S4 Table). Given these results, we next investigated whether this region had been identified in genome scans of Neanderthal and Denisovan introgression into modern humans. No hit was found within the most probable candidate genomic regions for Neanderthal introgression [31–33]. However, two independent studies recently reported two overlapping candidate regions for Denisovan introgression in Melanesians [34] and Papua New Guineans [35], respectively (S23 and S24 Figs and S4 Table). The two most distant SNP positions where the archaic genomes were found to carry derived alleles shared with the most frequent allele combinations in present-day humans and coinciding with at least one of these introgression maps delimitate a region of 70.61 kb (chr4: 41,977,828–42,048,441; GRCh38).

Consequently, we further explored the haplotype structure of this putatively introgressed region using all the SNP variants in the CHB, CEU and YRI populations from the 1000 GP [36] together with those of Oceanian individuals available in the HGDP dataset [37] presenting a homozygous genotype in the Denisovan genome, and including the corresponding nucleotide positions in the three high-coverage Neanderthal genomes previously used (see details in Materials and Methods). The resulting set of 112 SNPs defined 4 archaic haplotypes and a total of 61 modern human haplotypes with two major allelic configurations differing at 66 SNP positions: htI, present in 53% of the YRI, 13% in CEU and 2% in CHB chromosomes, and ht II, found in 95% of the CHB, 67% of the CEU, and in 3% of the YRI chromosomes. Notably, the two most frequent haplotypes in Oceanians (ht LX and ht LXI found in 52% and 30% of the Oceanian chromosomes, respectively) only show 2 and 3 SNP differences when compared with ht II (S5 and S6 Tables and see haplotype network in S25 Fig). We then used haplostrips [38] to better explore the haplotype structure of this region (chr4: 41,977,828–42,048,441; GRCh38) and genetic differences between modern human or Neanderthal haplotypes and the Denisovan genome. Notably, most of the Oceanian, CHB and CEU chromosomes (but also 8 out of 216 YRI chromosomes that carry the derived allele at rs1047626) displayed between 11 and 13 differences to the Denisova. In contrast, while the three Neanderthal haplotypes present between 17 to 28 differences to the Denisovan genome, the most common haplotype in YRI displays a total of 78 changes (Fig 2 and S7 Table). Moreover, the most frequent allelic configuration in CHB and CEU contains 77 derived alleles shared with the Denisovan genome, while the most prevalent in YRI only presents 11 derived alleles shared with archaic humans (Fig 2A and S5 Table). Similarly, the Neanderthal haplotypes, which all present the ancestral allele for rs1047626, share a minimum of 72 derived alleles with the Denisovan genome. To gain further insight into the genetic structure of such high-frequency Denisovan-like haplotype in modern humans and understand its presence in up to 7% of the YRI chromosomes, we repeated the same analyses including 4 additional Sub-Saharan African populations from the 1000 GP (i.e., ESN, MSL, GWD, and LWK; S26 Fig and S8 Table) and extending the explored region (chr4: 41,905,811–42,146,424; GRCh37; S27 Fig and S9 Table). Visual inspection of the corresponding haplostrips plots confirms that the Denisovan-like haplotypes carrying the derived Val allele in ESN (found in 4% of the ESN chromosomes), MSL (11%), GWD (10%) and LWK (8%) present the same archaic-like haplotype structure (and length ~70,614 bp) found in CHB and CEU (see S26 and S27 Figs). By contrast, the African haplotypes first clustered to the Denisovan and Altai Neanderthal configurations but presenting the ancestral allele at rs1047626, display fewer shared derived alleles with the Denisovan genome (besides that of the rs1047626 position) and seem to present a much-dispersed pattern of private and shared derived alleles with both archaic genomes that expands outside the SLC30A9 region and appears reshuffled in other African haplotypes (see S27 Fig and S9 Table). As for the Denisovan-like modern human haplotypes containing the Met to Val substitution, since the probability of observing one haplotype of 70,614 bp in length shared by ancestry without being broken by recombination given the local recombination rate of the region (1.58 cM/Mb) and the time since divergence between modern and archaic humans (see details in Materials and Methods) is highly unlikely (p-value < 2.07 x 10⁻⁸ when considering a mutation rate μ = 1 x 10⁻⁹; S10 Table), we can reject an scenario of incomplete lineage sorting (ILS) for causing the unusual haplotype configuration observed around the derived allele at rs1047626.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Potential signatures of adaptive introgression at SLC30A9.

(A) Haplotypes along the putatively introgressed region (chr4: 41,977,828–42,048,441; GRCh38) clustered and sorted by increasing distance with the Denisova genome shown in the first row. Each column corresponds to a polymorphic position with a homozygous genotype in the Denisovan (112 in total; see details in S4 and S5 Tables), each row is a phased haplotype, colored according to the population of origin of the individuals shown in the top. Note that all sites with a maximum within-population minor allele frequency below 0.05 were removed. Black cells represent the presence of the derived allele and white cells the ancestral allele (see Materials and Methods). The grey box highlights the location of rs1047626. (B) Number of SNP differences that the Neanderthal and modern human haplotypes present to the Denisovan genome sorted by decreasing similarity as obtained by Haplostrips [38]. (C) Genomewide distribution of S prime introgression values in Europeans and East Asians when using either YRI (values obtained from [31]) or Mbuti as an unadmixed group. The red dot denoted the SLC30A9 region. (D) Joint distribution of the U (10%, 50%, 100%) and Q95 (10%, 100%) statistics in CEU and CHB when using either YRI or ESN as unadmixed group. Circles denote the U (10%, 50%, 100%) and Q95 (10%, 100%) values for the SLC30A9 region in each corresponding analysis, which are also shown in the inserted table together with the 99.9th and 99th genomewide percentiles.

https://doi.org/10.1371/journal.pgen.1010950.g002

The presence of the putatively introgressed haplotype in YRI may explain why previous scans of Denisovan introgression have not detected SLC30A9 as a candidate region for archaic introgression in Eurasians. Since Mbuti do not present the derived allele of the Met to Val substitution (S28 Fig) and probably have not been reached by the back to Africa migration [39,40], we next computed the S prime statistic [31] in the Europeans and East Asian samples available in the Human Genome Diversity Panel (HGDP-CEPH) [37], using the Mbuti as an outgroup, and detected the SLC30A9 as a candidate for introgression in both groupings (Fig 2C). Finally, we also investigated the scenario of archaic introgression in CHB and CEU with the Q95 and U statistics [41] using both the African YRI and ESN population as outgroups and considering the 10% outgroup cutoff to account for the known presence of this putatively introgressed haplotype in Africa (Fig 2D and S11 Table). In CHB, Q95_{YRI, DEN} (10%, 100%) and Q95_{ESN, DEN} (10%, 100%) detected unusual frequencies for the derived alleles that are shared with the Denisovan genome (and at <10% frequency in YRI or ESN, respectively) when considering the 99.9th percentile of the statistic computed genomewide; while in CEU, Q95_{YRI, DEN} (10%, 100%) and Q95_{ESN, DEN} (10%, 100%) were only significant when considering the 99th percentile (Fig 2D). Similarly, the U statistic indicated an excess of shared alleles with the Denisova at more than 50% frequency in CHB and CEU (while being found at <10% frequency in YRI or ESN, respectively) only when considering the 99th percentile (Fig 2D and S11 Table, and see joint U and Q95 distributions indicating the SLC30A9 region together with two previously described candidate regions for Denisovan introgression in CHB in S29 Fig for comparison). Thus, overall, the numbers and frequencies of shared Denisovan alleles found in CHB and CEU provide further suggestive evidence for a likely scenario of adaptive introgression in the SCL30A9 region.

Signals of positive selection in Africa

As signatures of positive selection had been previously reported for the most abundant haplotype around SLC30A9 in YRI [13], we also searched for putative candidate variants displaying unusual past allelic trajectories in Africa. Notably, several derived allele variants characterizing the African-like haplotype of SLC30A9 were found to show marginal genome-wide p-values for positive selection with Relate (S4 Table). After their corresponding functional annotation (S13–S16 Tables), we selected one of the strongest eQTLs (i.e., rs4861157 at chr4:42046074, GRCh38) that also presented one of the highest iSAFE values along the region for further analysis with CLUEs, which estimated a selection coefficient s = -0.00425 and a logLR = 16.2401 for an adaptive event occurring within the last 1,000 generations (i.e., 28,000 years before present) in the YRI population (Figs 1B and S30–S33). Although the genotype for rs4861157 was not available in the high coverage archaic genomes and therefore it was not included in the computation of the haplotype network, we checked that its derived allele (A) is found in all chromosomes presenting either the most prevalent haplotype in YRI (ht I) or their corresponding derived haplotypes in the network (S25 Fig). These results confirm a pattern of contrasting directional selection between African and non-African populations acting on two major differentiated Yin/Yang haplotype configurations at SLC30A9. Moreover, as expected, the derived allele of rs4861157 has high frequencies in all African populations from the 1000 GP and associates with higher SLC30A9 expression (S1B and S33 Figs).

Molecular phenotype for Met50Val

We then investigated whether the Met50Val substitution in ZnT9 could determine any further differential molecular phenotype besides that of gene expression to explain this case of adaptative (positive) selection outside Africa. We first verified the predicted protein structure of ZnT9 using the AlphaFold algorithm. We observed that the polymorphic amino acid residue affected by rs1047626 is located at the cytosolic N-terminus domain of the ZnT9 transporter (Fig 3A and 3B). It has been suggested that the N-terminus of the ZnT family modulates the transporter activity by facilitating either protein interactions or zinc binding [42,43]. Thus, we proceeded to analyze whether the Met50Val substitution could cause any differential phenotype on the ZnT9 protein expression levels, cellular location, or zinc homeostasis at cellular level. Accordingly, we used a heterologous expression system in HEK293 cells overexpressing each of the two ZnT9 forms resulting from the Met50Val substitution, carrying either the ancestral allele (ZnT9-50Met) or the derived allele (ZnT9-50Val). Our experiments showed no difference in protein expression between the ZnT9-50Met and ZnT9-50Val variants after their corresponding 24h transient expression (Fig 3C). We next characterized the localization of the two ZnT9-50Met and ZnT9-50Val variants. For that, we performed immunostaining in non-permeabilizing and permeabilizing conditions. The results, shown in Fig 3D, indicate that the ZnT9 N-terminus is located intracellularly and follows a reticular pattern independently of the overexpressed variant. We further explored the subcellular distribution of the variants co-transfecting the ZnT9 constructs with the FEMP probe, a double fluorescent reporter plasmid for the ER (in CFP) and the mitochondria (in YFP). Our colocalization analysis gave similar results for the two ZnT9 variants, both showing a higher colocalization degree with the ER compartment (S34 Fig). We then used super-resolution STED microscopy to confirm the presence of ZnT9 in mitochondria as described by others [21,22] (Fig 4). ZnT9-HA was co-stained with anti-TOM20 (Fig 4A), a mitochondrial outer membrane resident protein. Both ZnT9-50Met and ZnT9-50Val variants showed co-localization with mitochondria. Remarkably, by analyzing the mean distance between mitochondria (Fig 4B) and the relative area occupied by mitochondria in the cell (Fig 4C), we observed that the overexpression of ZnT9 generated mitochondrial clustering in comparison with non-transfected cells. To check whether the distance between ER and mitochondria might be altered we carried out FRET analysis using the FEMP probe together with the ZnT9 variants or empty vector. Our results showed no differences in any of the tested conditions (Fig 4D). Overall, this initial characterization revealed no differences in cellular localization for either variant but suggests that ZnT9 has a role at the ER-mitochondria interface.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Characterization of the structure and overexpression of ZnT9 in HEK293 cells.

(A) ZnT9 protein structure prediction by AlphaFold, showing 6 transmembrane domains and a cytosolic N-terminus, where the Met50Val substitution is indicated by a red arrow. (B) Protein sequence of the ZnT9 transporter with the Met50Val substitution in red and the predicted alpha-helices corresponding to transmembrane domains in blue. (C) Representative Western blot against HA and GAPDH in cells transfected with ZnT9-50Met, ZnT9-50Val, or an empty vector. (D) Representative pictures of the membrane and intracellular immunostaining of HEK293 cells expressing ZnT9-50Met and ZnT9-50Val isoforms (both in green). Nucleae were stained with DAPI (blue). Scale bar = 10 μm.

https://doi.org/10.1371/journal.pgen.1010950.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Subcellular localization of ZnT9 variants.

(A) Superresolution STED microscopy in cells transfected with ZnT9-50Met and ZnT9-50Val immunoassayed with anti-HA (green) and with the mitochondrial marker anti-TOM20. Scale bar = 10 μm. (B-C) Bar graph representing mean inter-mitochondrial distance (B) and relative mitochondria area (C) in cells transfected with ZnT9-50Met or ZnT9-50Val (n = 10). *** p<0.001 using Bonferroni test between conditions. (D) Bar graph measuring FRET between the endoplasmic reticulum and mitochondria transfected with an empty vector (control), ZnT9-50Met or ZnT9-50Val together with the FEMP probe (n = 46–65). See further statistical details in S1 Data.

https://doi.org/10.1371/journal.pgen.1010950.g004

We then studied the impact of ZnT9 overexpression on cellular zinc homeostasis, comparing cells transiently transfected with either of the ZnT9 variants or an empty vector. First, the mRNA expression of the different ZIP transporters was characterized to analyze whether the overexpression of ZnT9, a transporter reported to promote cytosolic zinc efflux, is somehow compensated for by a modification in a zinc importer expression (Fig 5A). Our results showed alterations in ZIP12 and ZIP14 expression, although in the case of ZIP12 CT values were above 30 and ZIP14 displayed a reduction in cells overexpressing either of the two ZnT9 variants. We then used flow cytometry with Zinquin to measure the cytosolic zinc content in basal conditions or with excess zinc for 30 min and observed no differences in cytosolic zinc handling between vectors in the two tested conditions (Fig 5B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Characterization of the impact on cellular zinc homeostasis of ZnT9 variant overexpression in HEK293 cells.

(A) RT-PCR comparing the expression of several ZTs in basal conditions in cells transfected with ZnT9-50Met, ZnT9-50Val, or an empty vector. 2^-(DDCT) plotted using GAPDH as the housekeeping gene (ZIP1, ZIP6, and ZIP7 n = 6; Rest of ZTs n = 3); ** p<0.01 versus control using t-test, # p<0.05 versus control using ANOVA with Bonferroni correction. (B) Evaluation of zinc content by flow cytometry using Zinquin in 10 μM and 100 μM of ZnSO₄ (n = 9); *** p<0.001 using t-test. (C) Evaluation of endoplasmic zinc content using an endoplasmic reticulum fluorescent zinc sensor (ER-ZapCY1) in basal and 100 μM ZnSO₄ conditions (n = 14–20). * p<0.05 using t-test, ** p<0.01 using t-test. (D) RT-PCR comparing the expression of several ZnT transporters in basal conditions in cells transfected with ZnT9-50Met, ZnT9-50Val, or an empty vector. 2^-(DDCT) plotted using GAPDH as the housekeeping gene (n = 5–6) ** p<0.01, *** p<0.001 using Bonferroni-corrected ANOVA. (E) Evaluation of zinc mitochondrial matrix content using Mito-cCherry-Gn2Zn incubating 40min with basal and 100 μM ZnSO₄ conditions (n = 9–12); ** p<0.01 using t-test between basal and 100 μM zinc conditions, # p<0.05 using t-test between transfection conditions. (F) Viability MTT assay in cells transfected with ZnT9-50Met, ZnT9-50Val, or an empty vector incubated for 24h at 0, basal and 100 μM ZnSO₄ conditions (n = 8–15) * p<0.05, ** p<0.01 using ANOVA with Bonferroni correction between transfection conditions, # p<0.05 using Bonferroni-corrected ANOVA between zinc conditions. See further statistical details in S1 Data.

https://doi.org/10.1371/journal.pgen.1010950.g005

Given the presence of the ZnT9 transporter in ER membranes, we measured the endoplasmic zinc content using ER-ZapCY1 probe. No major alterations were observed in cells transfected with ZnT9-50Val, ZnT9-50Met, or an empty vector in basal conditions. However, upon incubation with 100 μM external zinc medium, only the cells transfected with ZnT9-50Met, or an empty vector had an increased ER zinc content, indicating that the overexpression of ZnT9-50Val was modifying the ER zinc homeostasis (Fig 5C). When analyzing the RNA expression of the known ER zinc importers ZnT5, ZnT6 and ZnT7, we found that the overexpression of ZnT9-50Val causes a downregulation of ZnT5 and ZnT6 expression (Fig 5D). Considering that ZIP7 is the major zinc exporter in the ER, we further explored its activity in conditions of ZnT9 overexpression. After 24h of transient expression, the results showed no difference in ZIP7 protein expression between the ZnT9 variants (S35A Fig). We also measured the ER zinc content in the presence of a ZIP7 blocker, finding a strong increase in the fluorescence of the ER zinc reporter. However, differences between basal and 100 μM external zinc conditions were no longer observed, nor between ZnT9 variants and the empty vector (S35B Fig), indicating that in the absence of ZIP7 activity the ER suffers a zinc overload.

As ZnT9 has been shown to regulate mitochondrial zinc homeostasis [21,22], we also interrogated the functional relevance of the Met50Val substitution by measuring the zinc content in the intermembrane space and the mitochondrial matrix, using SMAC-mCherry-GZnP2 and mito-Cherry-GZnP2 probes, respectively. In the intermembrane space, overexpression of ZnT9-50Val resulted in a slightly higher zinc content than cells overexpressing ZnT9-50Met or an empty vector. However, zinc increased equally in all conditions upon incubation with 100 μM zinc (S35C Fig). In the mitochondrial matrix, ZnT9-50Val-transfected cells had a higher zinc content in basal conditions as well but that was not further affected by incubating with 100 μM zinc, contrary to cells overexpressing ZnT9-50Met, or an empty vector (Fig 5E). Given that mitochondrial zinc content is reported to depend on ER zinc transport [21,22], similar experiments were carried out, but this time blocking the ER zinc exporter ZIP7 for 30 min to accumulate zinc in this organelle. The result was an increase of zinc levels in the mitochondrial matrix of control cells but not in ZnT9 overexpressing cells (S35D Fig). Moreover, the accumulation of zinc in the mitochondrial matrix of ZnT9-50Val-transfected cells is not restricted to HEK293. We observed similar results in the myoblast cell line C2C12 (S35E Fig). As a summary, we have generated a schematic model with the main findings related with the impact ZnT9 variants on intracellular zinc homeostasis (S35F Fig). In order to gain further insight into the ZnT9 transport activity in the mitochondria, ZnT9 was silenced with siRNA and the zinc content of the mitochondrial matrix was measured. As the data showed that incubation with 100 μM zinc increases zinc levels despite silencing ZnT9, we discarded the possibility that ZnT9 is a major zinc importer in the mitochondria (S36A Fig). Our expression data confirmed the silencing strategy and showed no alterations in the expression of ZnT5, the zinc importer in the ER, when ZnT9 was silenced (S36B Fig). Finally, we also investigated the impact of the alteration of zinc homeostasis caused by the ZnT9 variants overexpression on mitochondrial metabolism. MTT experiments were carried out in cells incubated for 24h with 0, basal and 100 μM zinc content. Higher metabolic activity signal was found in ZnT9-50Val-overexpressing cells both in basal and zinc excess conditions compared with the control (Fig 5F). Considering that at basal conditions cell number was not altered after 24h transfection with different plasmids (S37 Fig), MTT assays are revealing differences in mitochondrial metabolism. The absence of difference in the 0 zinc media supports the idea that the impact on metabolism is due to a zinc-dependent process.

Exploring an adaptive phenotype for Met50Val

As three SNPs in the 3’ flanking region of the SLC30A9 gene (rs2880666, rs6447133, rs7659700) have been previously identified as QTLs influencing zinc content in the liver [23] and they are found in moderate linkage disequilibrium (0.66 < r² < 0.76) with rs1047626 (S17 Table), we examined whether rs1047626 could be associated with such a systemic phenotype. No significant differences in liver zinc content were detected between individuals carrying the A (n = 74) or G alleles (n = 212; p-value = 0.107), nor between the AA (n = 11), AG (n = 52), and GG (n = 80) genotypes at rs1047626 (p-value = 0.322). However, in agreement with the known effects of the surrounding nutriQTLs with which the alleles at rs1047626 associate, the homozygous individuals for the G allele showed a tendency towards higher zinc concentrations in the liver than those homozygous for the A allele, whereas the heterozygotes AG were associated with an intermediate zinc content (S18 Table and S38 Fig).

Additionally, we searched for GWAS annotations linked to any of the 170 SNPs in high linkage disequilibrium (r² > 0.8) with rs1047626. We found reported GWAS hits for helping behavior (rs1507086, rs2660319, rs11051), total PHF-tau SNP interaction (rs4861153), and one SNP (rs34215985) associated with multiple psychiatric phenotypes, such as anorexia nervosa, hyperactivity disorder, autism spectrum disorder, bipolar disorder, major depression, obsessive compulsive disorder, and schizophrenia (S4 Table). A phenome-wide association study (PheWAS) for rs1047626 retrieved seven entries with a corrected p-value ≤ 1.5 x 10⁻⁵ (0.05/3,302 unique traits), including traits within the psychiatric (i.e., neuroticism and major depressive disorder, being the derived G allele of the rs1047626 polymorphism the allele that increases the risk of both phenotypes), metabolic (i.e., impedance, with the G allele associated with lower impedance measures), activities (i.e., fish oil, and glucosamine uptake, with the G allele increasing the phenotype value) and skeletal (i.e., height, with the G allele decreasing the phenotype value) domains, respectively (S19 Table).

Signals of positive selection in the SLC30A9 region

The PopHuman genome browser (http://pophuman.uab.cat) [26] was consulted to search for signatures of recent positive selection along the SLC30A9 region. This browser contains a broad range of genomic metrics and neutrality tests for all 26 human populations of the 1000 Genomes Project (1000 GP) [36]. Window-based computed values for F_ST, XP-EHH, and Fay and Wu’s H statistics were extracted from three geographically differentiated populations from Africa, Europe, and Asia: the Yoruba from Ibadan (YRI), Utah residents with Northern and Western European ancestry (CEU), and Han Chinese from Bejing (CHB). Gene annotations surrounding the SLC30A9 region were obtained from the Ensembl genome browser (http://grch37.ensembl.org/). The collected information was then represented using the Gviz package [58]. For each selection statistic and population, deviations from neutrality were considered as those greater than two standard deviations from the genomic mean.

Analysis of putative adaptive variants in SLC30A9

All SNPs for the longest SLC30A9 gene transcript (ENST00000264451) were extracted from the Ensembl Genome Browser (GRCh37 assembly). Only biallelic variants with a MAF ≥ 0.02 were kept for subsequent analysis and annotated using ANNOVAR [59] to obtain their corresponding genomic location, SNP type classification (i.e., coding, non-coding, synonymous, non-synonymous), and allele frequency information. All compiled variants were explored for associations in the GWAS Catalog (v1.0) and eQTLS in the GTEx Portal Dataset (V7) and they were further annotated with different in silico function predictors such as the score of Combined Annotation-Dependent Depletion (CADD score) [27], the Eigen score [60], and the FitCons score [61]. SNPs in high linkage disequilibrium (r²>0.8) with rs1047626 in CEU or CHB and their corresponding allele frequencies by continental region (AMR, EAS, EUR, SAS and AFR) were extracted from Ensembl. We run iSAFE [28] using a 300 kb region centered on rs1047626 (chr4: 41,853,671–42,153,671; GRCh37) from the CHB, CEU and YRI phase 3 1000 GP sequencing data and inferring the corresponding human ancestral allelic states from the Ensembl EPO multiple alignments.

Analysis of positive selection on candidate variants

We used Relate (https://myersgroup.github.io/relate) [29] to test whether the derived alleles at rs1047626 and rs4861157 were detected as fast spreading variants along multiple lineages when compared to competing non-carriers. For that, we first obtained the Relate-estimated coalescence rates, haplotypes and genealogies available for the 1000 GP (kindly provided by Leo Speidel), removed all sample haplotypes not belonging to CEU, CHB or YRI (with RelateFileFormats—mode RemoveSamples) and annotated SNPs as recommended in order to extract trees (with RelateFileFormats—mode GenerateSNPAnnotations) and the inferred genealogies for rs1047626 and rs4861157 in CHB, CEU and YRI (with RelateExtract—mode SubTreesForSubpopulation—years_per_gen 28), which were then visualized (with RelateTreeViewMutation -m 1.25E-8). Subsequently, we ran CLUES [62] to estimate the corresponding selection coefficients and allelic trajectories in either CEU and CHB or YRI using the SampleBranchLengths output from Relate as obtained with the Inference.py with N 30.000. The same procedure was repeated to run CLUES on FIN, PEL, JPT, and PJL.

Introgression

Allelic states in the Neanderthal and Denisovan genomes for those SNPs displaying high linkage disequilibrium (r²>0.8) with rs1047626 in CEU and CHB were queried from the genotypes available in four high coverage archaic genomes (https://cdna.eva.mpg.de/neandertal/). We initially inferred as putatively introgressed the genomic region comprised between the two more distant SNP positions found at high linkage disequilibrium (r²>0.8) with rs1047626 in CEU and CHB where the archaic genomes carried derived alleles shared with the most frequent allele combinations in present-day humans while also coinciding with at least one of the Denisovan introgression maps available for Melanesians [34] and Papua New Guineans [35]. To further explore a scenario of archaic introgression, all SNP genotypes along this putatively introgressed region (hg 38, chr4: 41977828–42048441) were extracted for all the Oceanians and for the CHB, CEU, and YRI populations from the corresponding VCF files of the HGDP [37] and 1000GP [63] sequencing projects, merged, and phased with Shapeit4 [64] using the recombination maps available (https://github.com/eyherabh/genetic_map_comparisons). In turn, VCF files for the Denisova, AltaiNeandertal, Vindija33.19, and Chagyrskaya-Phalanx archaic genomes were downloaded (https://cdna.eva.mpg.de/neandertal/), lifted over to hg38, and merged. We then crossed all the Denisovan positions found with a homozygous genotype in the putatively introgressed region with the corresponding Neanderthal positions and phased genotypes across the YRI, CHB, CEU and Oceanian populations. Only three positions in the Altai Neanderthal were found as heterozygotic positions and were kept in the analysis. From this complete set of inferred modern and archaic haplotypes (65 in total defined by 112 polymorphic positions, see S5 Table), we then computed nucleotide pairwise distances and plotted the haplotype network using the ape and pegas R packages [65]. The probability of observing a segment of 70,614 pb in-length shared with archaic humans due to incomplete lineage sorting (ILS) was calculated following the same procedure as in Huerta-Sánchez et al. (2014) [66] considering the generation (g = 25 years) and split times as in Dannemann et al. (2016) [67] and the average recombination rate of the putatively introgressed SLC30A9 segment (r = 1.58 x 10⁻⁸; extracted from https://github.com/eyherabh/genetic_map_comparisons). Briefly, we first calculated which is the expected length (L) for a segment shared by ILS using the formula L = 1/(r(t1+t2)/g) considering the split times for the divergence between humans and hominids (t1) and between Denisovan and Neanderthals (t2) given in Prüfer et al. (2014) [68] when considering two different mutation rates. Subsequently, the probability of observing an archaic segment of 70,614 bp in-length that persists due to ILS was computed as 1- GammaCDF (observed length, k, I/L), being 1/L the rate of the gamma distribution, L the expected length and k, the shape (k = 2) (see details in S10 Table). S prime was computed pooling all the whole genome sequences for the European and East Asian populations available at the HGDP dataset [37], respectively, and considering Mbuti as unadmixed outgroup. Singletons were excluded from the S prime analysis, and we only used biallelic sites without missing variants. S prime values estimated for CHB and CEU were directly extracted from Browning et al. (2018) [31]. U and Q95 statistics [41] testing for archaic introgression in CHB and CEU (phase 3 1000 GP sequencing data) were computed twice in the form U (10%, 50%, 100%) and Q95 (10%, 100%) using either YRI or ESN as outgroup (phase 3 1000 GP sequencing data), respectively, and considering only biallelic sites with no missing data. Significance was evaluated considering the 99th and 99.9th percentiles obtained from a genomewide distribution of the corresponding U and Q95 statistics computed in 40 kb windows as in Racimo et al. (2017) [41]

GWAS analysis

We used the GWAS catalog (https://www.ebi.ac.uk/gwas/) [69] and the PheWAS option in the GWAS atlas (https://atlas.ctglab.nl/) [70] to look for trait associations with rs1047626 and SNPs in high LD (r² >0.8). In the PheWAS analysis, a total of 3,302 unique traits as available at the GWAS Atlas database (last accession on 16/05/2022) were considered for Bonferroni multiple test correction (FDR = 1.5 x 10⁻⁵).

Generation of polymorphic sites

The plasmid containing the human SLC30A9 gene codifying for the ZnT9 transporter with an N-HA tag was obtained from Sino Biological Inc. (Catalog nr. HG22621-NY). The Met50Val polymorphism was generated via site-directed mutagenesis following standard conditions (QuikChange Lightning; Agilent) with the following primers: forward primer (5’-GACATTTGGAAGCTTTTCAAACGTGGTTCCCTGTAGTCA-3’) and reverse primer (5’-TGACTACAGGGAACCACGTTTGAAAAGCTTCCAAATGTC-3’). Before being used for cell transfection, both ZnT9-50Met and ZnT9-50Val isoforms were confirmed by sequencing. The LightRun sequencing service of Eurofins Genomics was used following the standard conditions for purified plasmid DNA samples with the following primers: forward primer 5’-GGCACAGAACTCAAAGCT-3’ and reverse primer 5’-TCTTCATGGGGACTTCGT-3’.

Genotyping of DNA liver samples

DNA samples and information regarding zinc concentration in the liver were obtained from the project of Engelken et al. (2016) [23], which determined micronutrient concentrations in liver samples from healthy individuals of western European origin, originally gathered by the UKHTB (for details, see Engelken et al. (2016) [23]). DNA concentrations for 143 available samples from the original study were quantified using a NanoDrop One spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). A genomic region of 226 bp comprising the rs1047626 position was amplified, purified, and sequenced using the LightRun sequencing service (Eurofins Genomics) following standard conditions for purified PCR products. The primers used for amplification were: forward primer 5’-GAAGCAGTGAAACACCTCTGG-3’, reverse primer 5’-TGTTTGTGATCCCTGTCCTTC-3’. For sequencing we used primer 5’-TGCAGCTAGGACTTGGTTTG-3’.

Cell culture and transfection procedure

HEK293 and C2C12 cells were cultured in DMEM supplemented with 10% FBS, 1% L-glutamine, and 1% penicillin and streptomycin (Basal medium) at 37°C in a humidified 5% CO₂ atmosphere. ZnSO₄ was added as needed to the final medium to generate specific Zn²⁺ concentration conditions. Cells were transiently transfected with ZnT9-50Met, ZnT9-50Val, pEGFP, or empty pCDNA3 vectors depending on the experiment. Polyethyleneimine (PEI) was used as the transfection reagent using 3 μg DNA for a 6-well plate or 1 μg DNA for a 24-well plate. In RNA interference experiments we used siRNA control (1027310, Qiagen) and siRNA ZnT9 (114789, Eupheria Biotech), and the transfection reagent was Lipofectamine 3000. Cells were incubated with the transfection solution for 3 hours, then washed and replaced with normal media. The experiments were performed 24h after transfection. To characterize the genotype of cells used for all the experiments, the DNA of HEK293 cells was extracted and purified following the NucleoSpin Tissue protocol (Macherey-Nagel, Düren, Germany). The genotype at rs1047626 was verified to be heterozygote (AG) in HEK293 cells with the same sequencing procedure previously described for the liver DNA samples.

Western blotting

Transfected cells were grown in 6-well plates and incubated for 24 hours. Cell lysis to detect ZnT9 protein was performed with 30 μL of lysis buffer containing 50 mM Tris-HCl at pH 7.4, 150 mM NaCl, 0.5% Nonidet P-40, and EDTA-free protease inhibition cocktail (Roche). Cell lysates were vortexed for 30 min at 4°C and centrifuged at 10,000× g to remove aggregates and boiled for 5 min at 95°C to lastly be placed on ice for 1 min. After electrophoresis in 12.5% polyacrylamide gel, proteins were transferred to nitrocellulose membranes using the iBlot system (Invitrogen, Waltham, MA, USA). Membranes were blocked with either 5% milk in TBS-Tween 0.1%, for GAPDH, or 5% BSA for HA-ZnT9, for 1 hour at room temperature. Primary antibodies were diluted in the corresponding blocking solution anti-HA for ZnT9 (H3663, Sigma-Merck), and anti-GAPDH (ab8245, Abcam). HRP secondary antibodies (1:1000; GE Healthcare) were used depending on the primary antibody. The ChemiDoc XRS+ system (Bio-Rad, Hercules, CA, USA) was used to obtain high-quality images. Quantity One Software (Bio-Rad) was used to analyze the results.

Immunostaining

For immunodetection of the two isoforms of ZnT9, for both cell surface and total cell expression, immunostaining was performed 24 hours after transfection with ZnT9-50Met or ZnT9-50Val plasmids in 24-well plates with coverslips coated with collagen. For cell surface expression, cells were incubated with anti HA (1:1000) in DMEM for 1h at 37°C. All samples were fixed with 4% paraformaldehyde (PFA) and only for total cell expression experiments were cells permeabilized with 0.1% Triton x-100 in PBS for 10 minutes. For all experiments, samples were blocked overnight with 1% BSA and 2% FBS in PBS. For total cell expression experiments, samples were incubated for 1h with an anti-HA antibody (1:1000 in blocking solution). For all experiments, cells were incubated with the secondary antibody (1:2000 in blocking solution), a goat anti-mouse Alexa Fluor 488 (Molecular Probes). Images were acquired using an inverted Leica SP8 Confocal Microscope with a 63× Oil objective and analyzed using ImageJ software.

In the colocalization analysis, to determine the subcellular distribution of the variants, immunostaining was performed 24 hours after transfection in 24-well plates with coverslips. Cells were co-transfected with ZnT9-50Met or ZnT9-50Val plasmids and a modified FRET-based ER-mitochondria proximity probe named FEMP [71]. The immunostaining procedure was the same as the one described above, except for the secondary antibody, which in this case was a goat anti-mouse Alexa Fluor 555 (Molecular Probes). The analysis of colocalization of ZnT9 with either the ER or mitochondria was performed with both Pearson’s correlation coefficient and Mander’s overlap coefficient using ImageJ software.

For STED imaging, cells were grown and seeded on 1.5H-thickness cover glasses. Primary antibodies were diluted in 2% BSA blocking buffer reagent. We used the rabbit anti-human TOM20 and the rabbit anti-human Calreticulin antibodies (1:1000) and secondary antibodies Abberior STAR RED or ORANGE (1:350). STED images were taken with a commercial Leica TCS SP8 STED 3× microscope equipped with a pulsed supercontinuum white light laser excitation source, using a 100× 1.4 NA oil HC PL APO CS2 objective. Post-analysis was performed on ImageJ. To measure the area occupied by the mitochondria with the Freehanded selection tool, we measured the cell region on the HA channel, subtracted the nucleus area and selected the area occupied by these organelles. The distribution of the mitochondria was quantified tracing a segmented line that connected all single mitochondria in one cell and then compare the length of these lines against the length in control cells.

Real-Time RT PCR

Cells transfected with either ZnT9-50Met, ZnT9-50Val, or empty pCDNA3 vectors were grown in 6-well plates and incubated for 24 hours. Total RNA was extracted from cells using Macherey-Nagel total RNA extraction kit. RNA was measured using a NanoDrop 1000 spectrophotometer (Thermo Fisher). cDNA was generated using a SuperScript Reverse Transcriptase system (Invitrogen). Quantitative PCR was performed using SYBR Green (Applied Biosystems) in the QuantStudio 12K system (Applied Biosystems). Primers are listed in S20 Table.

Zinc measurements

For the detection of zinc at the cytosolic level, cells were co-transfected with ZnT9-50Met or ZnT9-50Val vectors and pEGFP in 24-well plates. 24 hours after transfection, cells were incubated with 25 μM of Zinquin (Sigma-Aldrich, Darmstadt, Germany) for 30 min at 37°C (5% CO₂) in isotonic solution (ISO) containing 140 mM NaCl, 2.5 mM KCl, 1.2 mM CaCl₂, 0.5 mM MgCl₂, 5 mM glucose, and 10 mM Hepes (300 milliosmoles/liter, pH 7.4), and different concentrations of Zn²⁺. Cells were then dissociated with Trypsin 0.05% in 0.53 mM EDTA and washed with PBS. Fluorescence was quantified using an LSRII flow cytometer. Further analysis was performed using Flowing software to quantify Zinquin fluorescence of live transfected cells (Perttu Terho, Turun yliopisto, Turku, Finland).

To determine ER zinc levels, in vivo confocal imaging was used in cells co-transfected with the FRET-based ER-ZapCY1 probe (Addgene, Catalog nr. 36321)[72] and ZnT9-50Met, ZnT9-50Val, or empty pCDNA3 vectors in 6-well plates with 22 mm coverslips. 24 hours after transfection, cells were incubated with basal medium or supplemented with 100 μM ZnSO₄ for 3 hours. Then, samples were placed under the microscope in ISO for imaging with an SP8 Leica microscope (Wetzlar, Germany). When blocking ZIP7 activity, 1 μM of the ZIP7 inhibitor NVS-ZP7-4; HY-114395, MedChemExpress (Quimigen) or 1 μM of DMSO used as a control were added at 1:1000 in the treatment 3 hours before imaging. The measured zinc content is dependent on the FRET signal, which is expressed as the mean of YFP/CFP ratios from each condition normalized across experiments to an empty vector in basal zinc conditions. Images were analyzed using ImageJ software.

Mitochondria zinc concentration was measured in vivo using a plate reader (VICTOR Nivo, Perkin Elmer). Cells were co-transfected with Mito-cCherry-Gn2Zn or SMAC-Gn2Zn probes [51] and ZnT9-50Met, ZnT9-50Val, or empty pCDNA3 vectors in 24-well plates. 24 hours after transfection, cells were incubated for 40 min with basal medium or supplemented with 100 μM ZnSO₄. Then the media was changed to ISO solution and samples were placed in the plate reader for measuring. When blocking the ZIP7 activity, 1μM of the ZIP7 inhibitor NVS-ZP7-4 or DMSO was used as a control. Zinc content is expressed as the mean of GFP/Cherry ratios from each condition normalized to an empty vector in basal zinc conditions.

MTT assays

Cells were transfected with ZnT9-50Met, ZnT9-50Val, or empty pCDNA3 vectors in 24-well plates and incubated for 24h with normal medium, supplemented with 100 μM ZnSO4 or Zn²⁺-free growth medium generated with Chelex 100 resin. Then, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reagent was added (0.5 mg/mL) for 2 h at 37°C. After that, the supernatant was removed, and cells were resuspended in 400 μl of DMSO. The absorbance was read at 570 nm.