Multi-signal versus haplotype-aware cis-eQTL fine-mapping: the binding constraint is LD and relatedness structure, not method class

Question: How do multi-signal and haplotype-aware fine-mapping methods for cis-eQTLs in plant and animal populations compare in their ability to resolve multiple independent regulatory variants per gene?

Answer

Taken together, the retrieved papers support a specific and somewhat deflationary position: multi-signal credible-set frameworks—not haplotype-aware models—are the only approach for which per-gene, multi-variant resolution has been demonstrated at population scale in both plants and animals, with maize resolving causal signals for 16,270 eGenes and 77.7% of those eGenes carrying at least two credible sets [10], and the sheep and cattle atlases fine-mapping 322,468 primary plus 114,017 non-primary effects [13] and 433,972 primary plus 161,428 non-primary effects [14] respectively. Haplotype-aware work in these same species groups has largely been directed at QTL and haplotype-trait discovery rather than at decomposing regulatory signals within a gene [16], and where haplotype and dosage representations have been directly compared in a highly inbred crop panel, the two were "nearly confounded" and delivered nearly identical long-term outcomes [11]. Haplotype-aware modelling therefore earns its keep not as a general upgrade on multi-signal fine-mapping but in three narrower regimes: composite or cross-population-reversing LD backgrounds [2], non-additive allelic-context effects between closely spaced variants [3], and phasing-dependent allele-specific testing [4]. For animal populations specifically, the evidence indicates that relatedness- and LD-misspecification—not the absence of haplotype structure—is the dominant error source, since standard multi-signal tools lose accuracy outright when unrelatedness is assumed [20].

Multi-signal fine-mapping has already established pervasive per-gene allelic heterogeneity in crops and livestock

Maize leaf transcriptomes from 631 inbred lines yielded 17,468 eGenes with 16,270 fine-mapped, and the majority of fine-mapped eGenes were haplotype-rich, meaning they carried at least two credible sets; modelling all credible sets raised median R² from 0.16 to 0.27, and genes with more signals had higher expression heritability [10]. The same architecture appears in animals at atlas scale, where SheepGTEx reports 322,468 primary and 114,017 non-primary effects and explicitly attributes them to a high degree of regulatory allelic heterogeneity [13], and CattleGTEx Phase 1 reports 433,972 primary and 161,428 non-primary regulatory effects across seven molecular phenotypes, resolving 75% of GWAS signals for 44 traits [14]. Secondary signals are not merely residual noise: in TOPMed, 7,096 of 10,611 colocalising GWAS signals colocalised with at least one secondary e/sQTL signal, and secondary discovery remained unsaturated even where eGene discovery had saturated [7]. Method development has tracked this, with Bayesian multi-context fine-mapping that explicitly disentangles LD confounding outperforming both SuSiE and mvSuSiE [1].

Haplotype-based methods in plant and animal populations have targeted QTL discovery, not per-gene signal decomposition

The most developed haplotype-aware framework in crops defines LD-based haploblocks and computes local genomic estimated breeding values, improving QTL discovery and phenotypic prediction relative to single markers in barley [16]. Its objective, however, is locus detection and prediction, not enumeration of independent regulatory variants per gene. A direct comparison of phased-haplotype versus unphased-dosage fitness functions in a 444-line wheat nested association mapping population found only 18 of 30 selected founders in common and nearly identical long-term genetic gain, with the authors attributing the minimal difference to high homozygosity, where localGEBV and haplotype effects are nearly confounded [11]. Read-scale haplotype calling has been advanced for polyploids on the argument that haplotypes reduce dimensionality and increase statistical power [18], but the demonstrated outputs are genetic maps and diversity structure rather than cis-eQTL credible sets [18].

Haplotype-aware modelling decisively outperforms SNP-level resolution in composite and long-range LD backgrounds

Decomposing the MICA rs2596542 signal by carrier-set topology and non-negative matrix factorization showed that the variant is not a single regulatory proxy but a composite tag capturing separable MICA- and HLA-B/HLA-C-regulatory haplotypic backgrounds, with one axis population-stable and another enriched for signed-LD reversal [2]. In livestock, LD structure is actively hostile to SNP-proximal inference: analysis of chicken QTL regions found fragmented and interdigitated LD blocks, with candidate causal variants potentially far from the significant marker and even inside another QTL region [9]. Phasing also gates allele-specific readout: a joint haplotype-inference framework with phasability-aware testing avoids inflated false positives when haplotype information is incomplete, and resolves gene-, isoform-, and local-event-level cis effects missed by sequential variant-calling-then-phasing workflows [4].

Non-additive allelic context is the strongest mechanistic case for haplotype-aware models, and the strongest caution against single-variant credible sets

Massively parallel testing of 7,285 close-proximity SNV pairs, with at least one member annotated as an eQTL, found that 57% of pairs had a derived haplotype differing significantly in activity, that individual variant effects frequently depended on the neighbouring allelic background, sometimes with opposite sign, and that 59% of high-confidence pairs were non-additive, with non-additive pairs lying closer together [3]. Single-variant credible sets are structurally unable to represent such context dependence, which argues that haplotype-explicit parameterisation should matter more as variant density in credible sets increases.

In animal populations, relatedness-adjusted multi-signal fine-mapping is the necessary baseline against which any haplotype gain must be measured

Standard fine-mapping assumes unrelatedness and therefore performs poorly in livestock; a Bayesian framework built on a linear mixed model with a relatedness-adjusted LD matrix, plus adjusted FINEMAP and SuSiE variants, substantially outperformed FINEMAP, SuSiE, FINEMAP-inf, SuSiE-inf and GCTA-COJO in simulations on pig genotypes, with up to several-fold AUPRC improvements [20]. Because apparent cross-cohort and cross-ancestry differences in eQTL detection are driven by sample size, minor allele frequency and LD [5], and because sample sizes of 500 detect anywhere from under 0.1% to 60% of eQTLs depending on signal strength, with 2,000 samples reaching only 36.8% of all eQTLs [8], reported differences in the number of independent variants resolved per gene partly reflect power and LD rather than model class.

Conflicts and gaps

No retrieved paper performs a head-to-head benchmark of haplotype-aware versus multi-signal fine-mapping on the same cis-eQTL data in a plant or animal population; the comparison above is assembled across literatures with different endpoints. The inbreeding-dependence claim is inferred from a single wheat comparison [11] set against a maize result showing haplotype-rich credible-set architecture [10], and no study manipulates inbreeding level to test it. Fine-mapping in every retrieved atlas rests on SNP and short-indel panels, yet structural variants in perfect LD with the lead small variant raise SV enrichment among cattle eQTLs roughly sixfold [15], so credible sets may assign causality to the wrong variant class and haplotype-aware models built on SNPs inherit that error. The multi-ancestry fine-mapping methods with shared and ancestry-specific effects [21], continuous shrinkage priors with FDR control [19], and environment-aware multi-causal resolution [17] have been evaluated on human GWAS, not on cis-eQTLs in plants or livestock. Context-dependent effects that linear multi-signal models cannot see—quantile-specific pig eQTLs absent from standard linear regression and enriched in distal regulatory elements [12]—remain outside both method families.

What would change the answer

Applying haploblock-level and phased-haplotype parameterisations and SuSiE-type multi-signal fine-mapping to identical variants, samples and LD matrices within SheepGTEx, CattleGTEx and PigGTEx, with truth sets from haplotype-based simulation, would convert the current indirect comparison into a direct one [13] [14]. Building phased, SV-inclusive variant panels from long-read assemblies and repeating both fine-mapping families would test whether SNP-only credible sets understate haplotype-aware gains [15]. Allelic-series MPRA measurements designating which variant in each pair is causal, and whether the pair is additive, would supply the ground truth currently missing for evaluating either family [3]. Finally, stratifying the comparison by inbreeding and LD-block length—maize inbreds versus F1 hybrids, divergent cattle breeds, the chicken advanced intercross [6]—would test the prediction that haplotype-awareness pays off in proportion to how badly haplotypes and dosages are confounded.

References

[1] Lin, C., Lin, Y., Li, W., Xu, L., Zhang, X., Zhao, H.. "Leveraging cell-type specificity and similarity improves single-cell eQTL fine-mapping". bioRxiv. 2025-03-11T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.05.641709

[2] Ichikawa, Y.. "Carrier-set topology reveals composite MHC tagging at the MICA rs2596542 locus: insights from a flip-flop association example". bioRxiv. 2026-06-03T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.15.725353

[3] Kreevan, R., Pankratov, V., Yermakovich, D., Shukla, N., Metspalu, M., Eriksson, A., Dannemann, M., Gallego Romero, I., Org, T.. "Massively parallel characterization reveals context-dependent and non-additive regulatory effects of closely spaced variant pairs". bioRxiv. 2026-09-11T00:00:00.000Z. URL: https://doi.org/10.64898/2026.09.10.750584

[4] Xu, Z., Wang, K.. "LongAllele: a joint inference framework for allele-specific analysis on long-read bulk and single-cell RNA sequencing". bioRxiv. 2026-06-09T00:00:00.000Z. URL: https://doi.org/10.64898/2026.05.05.722992

[5] Gibbs, P. M., Beasley, I. J., Del Azodi, C. B., McCarthy, D. J., Gallego Romero, I.. "Power and linkage disequilibrium differences are major confounders of human eQTL portability across ancestries and cohorts". bioRxiv. 2026-07-26T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.26.708346

[6] Zhu, X., Zhai, C., Li, C., Liu, R., Mou, H., Zhu, Y., Luo, W., Chen, P., Wu, H., Wang, Y., Shi, K., Gong, M., Zheng, W., Ji, J., Luo, C., Qu, H., Shu, D., Hu, X., Fang, L., Wang, Y.. "Multi-layered molecular regulation of complex traits in an advanced intercross line". bioRxiv. 2026-08-21T00:00:00.000Z. URL: https://doi.org/10.64898/2026.08.17.745374

[7] Orchard, P., Blackwell, T. W., Kachuri, L., Castaldi, P. J., Cho, M. H., Christenson, S. A., Durda, P., Gabriel, S., Hersh, C. P., Huntsman, S., Hwang, S., Joehanes, R., Johnson, M., Li, X., Lin, H., Liu, C.-T., Liu, Y., Mak, A. C. Y., Manichaikul, A. W., Paik, D., Saferali, A., Smith, J. D., Taylor, K. D., Tracy, R. P., Wang, J., Wang, M., Weinstock, J. S., Weiss, J., Wheeler, H. E., Zhou, Y., Zoellner, S., Wu, J. C., Mestroni, L., Graw, S., Taylor, M. R. G., Ortega, V. E., Johnson, C. W., Gan, W., Abecasis, G., Nickerson, D. A., Gupta, N., Ardlie, K., Woodruff, P. G., Zheng, Y., Bowler, R. P. "Cross-cohort analysis of expression and splicing quantitative trait loci in TOPMed". medRxiv. 2025-02-21T00:00:00.000Z. URL: https://doi.org/10.1101/2025.02.19.25322561

[8] Rosen, J. D., Broadaway, K. A., Brotman, S. M., Mohlke, K. L., Love, M. I.. "Higher eQTL power reveals signals that boost GWAS colocalization". bioRxiv. 2025-08-05T00:00:00.000Z. URL: https://doi.org/10.1101/2025.08.05.668745

[9] Lipkin, E., Fulton, J. E., Smith, J., Burt, D. W., Soller, M.. "Is it chaotic? Chicken Linkage Disequilibrium and quantitative trait loci". bioRxiv. 2025-09-29T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.28.679118

[10] Grzybowski, M. W., Schnable, J. C.. "Haplotype-rich cis-regulation underlies transcriptomic diversity across the breeding history of maize (Zea mays)". bioRxiv. 2026-10-19T00:00:00.000Z. URL: https://doi.org/10.64898/2026.02.19.706772

[11] Shaffer, W., Papin, V., Carter, Z., Brunner, S. M., Tong, J., Villiers, K., Robinson, H., Voss-Fels, K., Hayes, B. J., Hickey, L., Dinglasan, E.. "Comparison of localGEBV and Optimal Haplotype Stacking Fitness Functions using a Novel R Package: HapSelect". bioRxiv. 2026-07-13T00:00:00.000Z. URL: https://doi.org/10.64898/2026.07.08.737160

[12] Wang, F., Wang, C., Teng, J., Fang, L., Ionita-Laza, I.. "Context-dependent genetic regulation of gene expression in pigs". bioRxiv. 2026-03-13T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.11.710275

[13] Gong, M., Zhuang, Z., Sun, X., Xu, Y., Zhang, H., Wang, Y., Guan, D., Li, R., Lu, X., Bai, Z., Feng, P., Song, M., Tian, M., Lu, J., Wang, M., Lu, X., Wu, D., Su, P., Liu, P., Zhang, G., Shi, J., Shan, M., Zhang, Y., Meng, Z., Li, H., He, X., Yang, J., Song, Y., Li, X., Du, X., Zhang, X., Yang, H., Teng, J., Li, H., Zhu, X., Zhang, H., Lin, Q., Zhu, D., Lin, B., Liu, X., Liu, J., Zheng, W., Gong, W., An, B., Zhang, Q., Sahana, G., Lund, M. S., Li, C., Guo, J., Wang, X., Liu, Y., Zhao, B., Yao, X., Zhang, Y., Wang, F., Zheng, W., Huang, J., Wang, S., Di, J., Tulafu, H., Liu, Z., Fu, S., Liu, Y.. "A multi-tissue atlas of genetic regulatory effects in sheep". bioRxiv. 2026-06-02T00:00:00.000Z. URL: https://doi.org/10.1101/2025.11.03.686346

[14] Li, H., Zhang, H., Zhu, D., Zhao, P., Wei, Z., Lu, J., Gong, M., Zhang, Q., Zheng, W., Liu, X., GUAN, D., Teng, J., Lin, Q., Tang, Y., Gao, Y., Zhao, S., Zhang, Z., Du, J., Fang, C., An, B., Lin, B., Zhang, H., Tian, M., Tian, J., Chen, S., Liu, W., Wang, Y., Wang, M.-S., Ibeagha-Awemu, E. M., Crooijmans, R., Derks, M., Godia, M., Madsen, O., Pausch, H., Leonard, A. S., Frantz, L., MacHugh, D. E., Grady, J. F. O., Ionita-Laza, I., Zhao, X., Guan, L., Zhou, H., Marmol-Sanchez, E., van der Wijst, M., Lu, X., Jiang, H., Yang, Z., Yang, Q., Liu, Q., Xu, C., Li, M., Hou, Y., Pan, Z., Chen, Y., Xian. "The enhanced multi-tissue atlas of regulatory effects in cattle". bioRxiv. 2026-03-21T00:00:00.000Z. URL: https://doi.org/10.64898/2026.03.18.712441

[15] Mapel, X. M., Leonard, A. S., Pausch, H.. "Molecular QTL are enriched for structural variants in a cattle long-read cohort". bioRxiv. 2025-10-13T00:00:00.000Z. URL: https://doi.org/10.1101/2025.05.16.654493

[16] Shaffer, W., Papin, V., Yadav, S., Voss-Fels, K. P., Hickey, L., Hayes, B., Dinglasan, E. G.. "Local genomic estimates provide a powerful framework for haplotype discovery". bioRxiv. 2026-05-26T00:00:00.000Z. URL: https://doi.org/10.1101/2025.08.28.672830

[17] Wang, S., Ojewunmi, O., Pirie, F., Motala, A., Ramsay, M., Morris, A., Fatumo, S., Chikowore, T., Asimit, J. L.. "Multi-ancestry fine-mapping accounting for ancestral and environmental heterogeneity improves resolution and interpretation". bioRxiv. 2025-11-28T00:00:00.000Z. URL: https://doi.org/10.1101/2025.11.26.690727

[18] Rio, S., Abdallah, S., Durand, T., DHont, A., Garsmeur, O.. "HaploCharmer: a Snakemake workflow for read-scale haplotype calling adapted to polyploids". bioRxiv. 2025-09-22T00:00:00.000Z. URL: https://doi.org/10.1101/2025.03.14.642807

[19] Li, X., Xiong, Z., Sham, P. C., Zhang, Y. D.. "MACHINE: a robust and scalable multi-ancestry fine-mapping method using a continuous global-local shrinkage prior". medRxiv. 2025-09-30T00:00:00.000Z. URL: https://doi.org/10.1101/2025.09.28.25336857

[20] Wang, J., Tiezzi, F., Huang, Y., Wei, J., Maltecca, C., Jiang, J.. "Fine-mapping methods for complex traits: essential adaptations for samples of related individuals". bioRxiv. 2025-06-15T00:00:00.000Z. URL: https://doi.org/10.1101/2021.10.10.463846

[21] Mirmira, T., Ma, N., Margoliash, J., Gonzalez Rivera, W. G., Amariuta, T., Frazer, K. A., Goren, A., Gymrek, M.. "Characterization of shared and ancestry-specific signals driving complex traits using multi-ancestry fine-mapping". medRxiv. 2025-11-15T00:00:00.000Z. URL: https://doi.org/10.1101/2025.11.13.25339614

How this survey was made

This survey answers one question: “How do multi-signal and haplotype-aware fine-mapping methods for cis-eQTLs in plant and animal populations compare in their ability to resolve multiple independent regulatory variants per gene?”

We keep our own collection of recent preprints and refresh it every day from bioRxiv, medRxiv, arXiv, PsyArXiv, SocArXiv and ChemRxiv. For each survey, we search that collection for recent work and keep the strongest matches: papers that use the same terms as the question, and papers on the same topic in different words.

The text was written from the titles and abstracts of those papers; the full texts were not read. Statements that draw on a particular paper are marked with a number in brackets. The reference list just above is the 21 papers whose work this text draws on, chosen from the 145 papers those searches returned; the rest of the shortlist is not printed. It is not a complete review of everything published on the question.

Treat this as a starting point and a reading list, not a peer-reviewed review. The searches behind this survey ran on October 21, 2026 and covered work from the previous two years.