Volunteers from Germany, Poland, UK, USA, Chile, India and South Africa continuously provide samples (buccal swabs) to DKMS for their registration as potential stem cell donors. Between 2017 and 2021, approximately 3.6 million samples were genotyped for MICA (Germany 56%; Poland 18%; UK 15%; US 8%; Chile 2%; India 1%; South Africa 0% (donor center not yet active)). This cohort was used to identify and characterize novel alleles. Another 93,814 samples were genotyped for MICA from 2023 to 2024 and used for MICA population frequency analyses (Germany 65%; Poland 13%; South Africa 8%, Chile 7%; UK 4%; US 3%; India 0.2%). As part of the registration process, the donors are asked to self-assign their ethnic background. All subjects gave written informed consent in accordance with the Declaration of Helsinki. The described genotyping is within the scope of the consent forms signed at recruitment.

Samples for the registration of potential stem cell donors are genotyped in an high-throughput workflow that targets HLA-A, -B, -C, -E, -DPB1, -DQB1, -DRB1, -DPA1, -DQA1, -DRB3/4/5, MICA and MICB (MICA/B), KIR, blood groups ABO and Rh, and CCR5 as described before (24–30). A detailed description of the workflow with a focus on MICA genotyping can be found in Klussmeier et al. (24). In brief, MICA exons 2, 3, 4, 5 are amplified by PCR (complete coverage of exons 2 and 3, partial coverage of exons 4 and 5). After pooling the PCR products with the HLA loci of the same donor, an indexing PCR is performed. Before 2019, the PCR products of up to 3,840 donors were pooled, cleaned up and sequenced using HiSeq Rapid SBS Kits V2 (500 cycles) on HiSeq2500 instruments (Illumina, San Diego, USA). After 2019, up to 7520 potential stem cell donors were sequenced using NovaSeq6000 SP Kits (500 cycles) on a NovaSeq6000 instrument (Illumina, San Diego, USA). Genotyping of high-throughput sequencing data was performed by neXtype (24, 25). Since not all bases of MICA are covered by our workflow, some genotyping results are ambiguous. Here, we report them by a representative allele, which is marked with a hash symbol (#) (Table 1). Previously, we described haplotypes with MICA duplications and MICA deletions (31). While neXtype correctly genotypes MICA duplications and reports three MICA alleles in such samples, it reports a homozygous instead of a hemizygous result for samples with MICA deletions. Nevertheless, since MICA deletions are rare (e.g., 0.3% in Europe, 2.5% in Chile), we accepted that this might minimally influence allele frequency calculations.

Samples with novel MICA alleles were subjected to two independent long-range PCRs (12 kB) that amplify the complete MICA gene from 5’ to 3’UTR. The following primers were used: CTGCTTGAGCCGCTGAGAGG (forward), GATCCAGGCAGGGAATTGAATCCC and GAGATCCAGGCAGGGAATTCAATTCC (reverse). In detail, 4 μL genomic DNA was combined with 0.08 μM primer mix, 1x Advantage Genomic LA Buffer, 1.25 U Advantage Genomic LA Polymerase Mix (Takara Bio, Mountain View, California), and dNTPs (0.4 mM each) in a 25 μL reaction volume. PCR conditions: 94 °C 1 minute, 35 cycles: 98 °C 10 seconds/65 °C 12 minutes, 72 °C 10 minutes. PCR success was checked by agarose gel electrophoresis. The product of one PCR reaction was used for Illumina shotgun sequencing as described before (32–34). In brief, fragmentation and adapter ligation was performed according to “NEBNext Ultra II DNA Library Prep Kit for Illumina” protocol (New England Biolabs, Ipswich, Massachusetts). After purification with 0.7x SPRIselect beads (Beckman Coulter, Brea, California), custom barcodes were attached by a 7-cycle-indexing PCR. Finally, 48 samples were pooled and subsequently purified using 0.7x SPRIselect beads. After qPCR library quantification, four pools (up to 192 samples) were sequenced on a MiSeq instrument using a MiSeq Reagent Kit v2 (500 cycles) according to the manufacturer’s instructions (Illumina, San Diego, California). The product of the second PCR reaction was used for SMRT sequencing (Pacific Biosciences, Menlo Park, California) as described before (32). PCR products of the prior long-range PCR were barcoded by an additional 10-cycle PCR reaction with indexing primers (0.2 μM). 192 samples were then pooled and library preparation was carried out according to the manufacturer’s instructions. Libraries were size selected with the BluePippin system using a 0.75% cartridge (Sage Science, Beverly, Massachusetts) and sequenced on a Sequel instrument using Sequel Sequencing Kit 3.0, SMRT Cell 1 M v3 and a 10 hour movie (Pacific Biosciences, Menlo Park, California).

Sequencing reads were analyzed using NGSengine (GenDx, Utrecht, The Netherlands) and dual redundant reference sequencing (DR2S) as described before (35, 36). The versions of the IPD-IMGT/HLA Database used in this analysis ranged from release 3.35 (2019) to release 3.48 (2022), with each sample batch analyzed using the most current version available at the time (refer to the analysis date of individual sequences in Supplementary Data). Samples with low sequencing quality and not fully conserved consensus sequences were discarded from analysis. Finally, all approved sequences were submitted to the IPD-IMGT/HLA Database using TypeLoader2 (37, 38). In general, all novel sequences were submitted. In addition, we submitted sequence extensions for alleles so far only partially described in the IPD-IMGT/HLA Database. Often, either was true for both alleles of a sequenced sample. If two identical sequences from different samples were available, the second sequence was submitted as confirmation.

In general, samples that failed in PCR and/or analysis were not repeated. We know from experience that this is usually caused by insufficient DNA quality, especially DNA fragmentation, and will not improve by repetition. To deal with this issue, three to five samples with the same targeted novel variation were selected for sequencing if enough samples were available.

MICA protein sequences were obtained from the IPD-IMGT/HLA Database (release 3.60) and aligned using CLC Genomics Workbench (version 24.0) (Qiagen Digital Insights, Aarhus, Denmark). Only sequences with complete amino acid coverage were included. A custom R script was used to compare every amino acid in the alignment to the corresponding amino acid of the reference allele MICA*002. Finally, alleles were sorted manually to generate clusters that visually highlight the similarity of alleles to the most frequent ones.

A distance matrix was calculated from the alignment using hamming distance and a neighbor-joining tree with midpoint rooting was built using the R package ape version 5.8.1 (39). Sequences without complete amino acid coverage and null alleles were excluded. Visualization was performed using the R package ggtree version 3.14.0 (40). For improved visualization, the branch lengths of the tree were square rooted before plotting the tree.

MICA population frequencies were calculated using samples that were genotyped in the high-throughput workflow with IPD-IMGT/HLA Database release 3.50 or higher. At this time (January 2023), all our submitted novel exon variations were officially named by the IPD-IMGT/HLA Database and consequently used for genotyping by neXtype. As part of the registration process as potential stem cell donors, the donors are asked to self-assign their ethnic background. These data were used for calculating MICA population frequencies. Since selectable ethnicities varied between the different DKMS donor center questionnaires, data were only grouped within one donor center (e.g., samples indicated as DE_Turkey were collected in Germany but the donor self-assigned to a Turkish ethnic background). Populations with more than 1,000 genotyped samples were selected for calculating MICA frequencies (DE_Germany, PL_Poland, ZA_Black, CL_Non-Indigenous, UK_British/Irish, ZA_White, DE_Turkey). Due to lacking sequence information outside of exons 2-5, MICA population frequencies were only calculated at protein resolution (first field). For samples with phasing ambiguities, the probability of each possible result was calculated based on the allele frequencies of unambiguously typed samples in the respective population. According to these probabilities, counts were added to the different alleles. Ambiguities that cannot be resolved by our workflow are listed in Table 1.

In 2017, we added MICA genotyping to our high-throughput stem cell donor workflow. At that time, 107 MICA alleles were listed in the IPD-IMGT/HLA Database, of which 92 (86%) were only partially described (release 3.35, January 2019). Because partial allele entries in the database can complicate genotyping, our initial goal was to extend the sequences of frequently observed partial MICA alleles in the IPD-IMGT/HLA Database. Hence, we selected 299 samples with partial sequence coverage and sequenced MICA in full-length. Thereby, each targeted allele was covered by multiple samples. After sequence analysis, we could successfully extend the sequences of 35 distinct, previously only partially described, MICA alleles. Overall, this first batch resulted in 209 sequence submissions to the IPD-IMGT/HLA Database, among them 22 alleles coding for novel MICA proteins, 9 synonymous exon variations, 70 intron variations and 108 confirmations/sequence extensions (Supplementary Data). These alleles were incorporated in the IPD-IMGT/HLA Database releases between January and October 2020. By release 3.42, the number of MICA alleles had increased to 224, of which 159 were described in full-length (71%) (Figure 1).

By 2020, we had genotyped approximately 3.6 million samples, of which 11,091 contained a novel sequence (0.3%) in exons 2, 3, 4, or 5. However, some of these novel sequences were observed repeatedly, e.g. the most frequent novel sequences were identified in 1,273 and 763 samples (these sequences were later named MICA*141 and MICA*119, respectively). Overall, we identified 1,103 distinct novel sequences of which 145 were detected more than ten times. In contrast, 559 variations were observed only once and are presumably very rare alleles.

For optimal use of our resources, we focused the second batch of novel MICA allele characterization on the 145 most frequent variants. A total of 474 samples were selected to cover each variation with multiple samples. Lower-frequency variations were added only to fill plates. As expected from prior experience of long-range PCRs on buccal swab derived DNA, approximately 33% of samples failed (25% in PCR, 8% in analysis) (34), likely due to DNA fragmentation. However, reasons for PCR failure were not further investigated for individual samples. Following analysis, this second batch of MICA novel allele characterization led to 394 sequence submissions to the IPD-IMGT/HLA Database, among them 177 alleles coding for novel proteins, 64 synonymous exon variations, 64 intron variations and 89 confirmations/sequence extensions. These include 139 (96%) of the targeted 145 frequent variations.

Combining both batches, we submitted 603 sequences to the IPD-IMGT/HLA Database: 199 novel proteins, 73 synonymous exon variations, 134 intron variations, and 197 confirmations/sequence extensions (Figure 1A). These sequences now represent approximately two-thirds of all MICA alleles listed in the IPD-IMGT/HLA Database (release 3.60, April 2025) (Figure 1B).

Among the characterized and submitted MICA alleles were 199 coding for novel MICA proteins. A detailed overview of all base variations in comparison to the closest known allele at the time of sequence submission is provided in Supplementary Data.

At first-field (protein) resolution, MICA*008, MICA*002 and MICA*009 have been identified as the most common alleles in the German population (24). Consequently, it is not surprising that more than half of the submitted novel alleles are variations of these alleles (Figures 2, 3). Nevertheless, we identified variations of all other frequent MICA alleles. Specifically, most of the previously undescribed amino acid variations appear to be randomly distributed within the regions covered by our high throughput genotyping workflow (amino acids 1–181 and 204–319 of the mature protein) (Figure 3).

The most extensively studied amino acid variation in MICA is the Met/Val dimorphism at position 129. Among the common alleles, MICA*002, MICA*007, MICA*011, MICA*012, MICA*017, and MICA*018 encode a methionine at this position, while MICA*004, MICA*008, MICA*009, MICA*010, MICA*016, MICA*019, and MICA*027 encode valine (Figure 3). Notably, two of our novel alleles are exceptions regarding this amino acid. While MICA*147 and MICA*202 are otherwise very similar to the valine-encoding MICA*008 (Figure 2), they encode methionine at position 129.

Additionally, we report five new MICA null alleles. Frameshift mutations in MICA*096N and MICA*107N are present in exon 2, while those of MICA*195N, MICA*222N, and MICA*286N are located in exon 3 (Figure 3; Supplementary Data). These mutations are predicted to result in non-functional proteins.

In 2020, we published MICA allele frequencies for the German population based on over one million samples, identifying novel alleles at a cumulative allele frequency of 0.3% (24). After characterization of the most frequent novel alleles, our next objective was to analyze the allele frequencies of the previously novel alleles across different populations.

Our independent new cohort consisted of 93,814 samples genotyped for MICA between 2023 and 2024 using our high-throughput workflow. Within this cohort, we identified seven populations with over 1000 samples each: DE_Germany (n=48,618), PL_Poland (n=11,776), CL_Non-Indigenous (n=4,937), ZA_Black (n=4,085), UK_British/Irish (n=2,090), ZA_White (n=1,989) and DE_Turkey (n=1,823). These samples were used to calculate allele frequencies at protein resolution (first field).

Our largest population, DE_Germany, confirmed the results from our previous study (Table 2) (24). The most frequent MICA allele was MICA*008 (44%), followed by MICA*002# (11%), MICA*009# (9%), MICA*010# (8%), and MICA*004 (7%). Among the previously novel MICA alleles, the most frequent alleles in the German population were MICA*107N (0.02%), MICA*141, MICA*089, MICA*119 and MICA*136 (all 0.01%) (Table 2; Supplementary Data).

In the Polish population, allele frequencies were largely similar to the German population. The British/Irish population showed the highest MICA*008 frequency (51%) among all studied populations (Table 2; Supplementary Data).

Interestingly, the South African White population exhibited MICA allele frequencies closely resembling those of the other European populations. This contrasts with the South African Black population. Even though MICA*008 remained the most frequent MICA allele, its frequency was only 28% (43% in ZA_White). MICA*004 had a notably higher frequency (24%) in the South African Black population than in any other studied population, followed by MICA*002# (22%), MICA*019 (5% vs. 0.8% in ZA_White), and MICA*015 (5% vs. 0.3% in ZA_White). Conversely, other MICA alleles were underrepresented in the South African Black population: MICA*010# (0.05% vs. 8% in ZA_White), MICA*007 (0.2% vs. 4% in ZA_White), and MICA*017 (0.03% vs. 3% in ZA_White) (Table 2).

In the non-indigenous Chilean population, MICA*002# (31%) was the most frequent MICA allele, followed by MICA*008 (19%) and MICA*004 (10%). In the Turkish population residing in Germany, MICA*008 (21%) was followed by MICA*009# (20%) and MICA*002# (14%). Notably, MICA*016 had a frequency of 10% in this group, compared to only 2% in the German population.

Some novel alleles appeared to be population-specific. MICA*258 (n=26) and MICA*008:28 (n=45) were almost exclusively detected in individuals from South Africa that self-assigned as Black or Colored. Only one individual with MICA*008:28 self-assigned an Indian ethnic background. MICA*244 was exclusively identified in individuals of the ZA_White population (n=8) and all individuals with MICA*004:02 self-assigned a Polish or Russian ethnic background (n=6).

As expected, the characterization and submission of the novel alleles substantially reduced the cumulative frequency of novel alleles from 0.3% to 0.03% in the German population. However, less sequenced populations such as ZA_Black and DE_Turkey still reported higher cumulative novel allele frequencies (0.1%). This was likely due to the underrepresentation of these populations in the workflow and therefore lower prioritization for novel allele characterization.

Overall, in this independent cohort of 93,814 samples, we reidentified 120 of the 199 submitted novel MICA proteins. The remaining 79 were not detected again and are presumed to be rare.

It is well known that MICA is in strong linkage disequilibrium to HLA-B (23). However, due to the absence of phased genotype data, we are unable to determine HLA-B linkage information for every novel MICA allele. For alleles identified in multiple samples, though, we could infer the most likely linkage. For example, MICA*107N was identified 57 times in the cohort that was used for MICA frequency calculations, and all samples were positive for an HLA-B*14:02:01G allele, as well. Consequently, based on this co-occurrence, we conclude that MICA*107N and HLA-B*14:02:01G share a haplotype. Similarly, Table 3 lists every novel MICA allele that was detected in at least 10 samples, of which all were reported with the given HLA-B allele.