Genetic evaluations and genome-wide association studies for specific digital dermatitis diagnoses in dairy cows considering genotype x housing system interactions

The present study aimed to use detailed pheno-typing for the claw disorder digital dermatitis (DD) considering specific DD stages in 2 housing systems (conventional cubicle barns = CON and compost bed-ded pack barns = CBPB) to infer possible genotype x housing system interactions. The DD-stages included 2,980 observations for the 3 traits DD-sick, DD-acute and DD-chronic from 1,311 Holstein-Friesian and 399 Fleckvieh-Simmental cows. Selection of the 5 CBPB and 5 CON herds was based on a specific protocol to achieve a high level of herd similarity with regard to climate, feeding, milking system and location, but with pronounced housing system differences. Five other farms had “a mixed system” with 2 sub-herds, one representing CBPB and the other one CON. 899 cows (1530 observations) represented the CBPB system, and 811 cows (1450 observations) the CON system. The average disease prevalence was 20.47% for DD-sick, 13.88% for DD-acute and 5.34% for DD-chronic, with a higher prevalence in CON than in CBPB. After quality control of 50K genotypes, 38,495 SNPs from 926 cows remained for the ongoing genomic analyses. Genetic parameters for DD-sick, DD-acute and DD-chronic were estimated by applying single-step approaches for single-trait repeatability animal models considering the whole data set, and separately for the CON and CBPB subsets. Genetic correlations between same DD traits from different housing systems, and between DD-sick, DD-chronic and DD-acute, were estimated via bivariate animal models. Heritabilities based on the whole data set were 0.16 for DD-sick, 0.14 for DD-acute and 0.11 for DD-chronic. A slight increase of heritabilities and genetic variances was observed in CON compared with the “well-being” CBPB system, indicating a stronger genetic differentiation of diseases in a more challenging environment. Genetic correlations between same DD traits recorded in CON or CBPB were close to 0.80,


INTRODUCTION
One of the most important claw disorders in dairy cattle with the highest incidence and pronounced effects on farm economy is digital dermatitis (DD) (Klitgaard et al., 2014;Solano et al., 2016).Digital dermatitis is a multi-factorial claw infection (Blowey and Sharp, 1988;Read and Walker, 1994).In addition to the predominant bacterial effects caused by the genus Treponema spp., housing conditions, feeding characteristics and genetics of the host play an important role (Döpfer et al., 2012;Solano et al., 2017).In some countries, official genetic evaluations for DD have been established, which are mostly based on binary disease diagnoses from producers and claw trimmers (Rensing, 2019).The pedigree based binary DD heritabilities were in a small to moderate range from 7.3% to 14.2% (König et al., 2005;Schöpke et al., 2015;van der Linde et al., 2010), and from 5% to 36.7% when considering genomic relationship matrices (Biemans et al., 2019;Shabalina et al., 2020a).Enlarging the data pool for genomic predictions and genetic parameter estimations is possible in so-called single-step approaches, simultaneously considering genomic and pedigree relationships (e.g., Lourenco et al., 2020;Misztal et al., 2020).With regard to functional traits, single-step evaluations have been carried out for milkability (Guarini et al., 2019a), for female fertility (Matilainen et al., 2018;Guarini et al., 2019b) and for diseases (Vukasinovic et al., 2017).Shabalina et al. (2020a) compared genetic parameter estimates for clinical mastitis considering the different matrices A, G and H, but results from pure genomic and single-step approaches were quite similar.
In genomic evaluations for claw disorders, mostly producer-recorded disease diagnoses or data from routinely claw trimming are used.The advantage of producer-recorded diseases is the utilization of simplified diagnosis schemes as mostly implemented in herd management software programs, contributing to large data sets for genetic evaluations (Zwald et al., 2004).However, for complex diseases and especially for DD with specific acute and chronic stages, disease pathogenesis is insufficiently depicted through simple binary diagnoses (Döpfer et al., 1997).Schöpke et al. (2015) introduced a recording system for different DD stages for quantitative genetic analyses.Such comprehensive DD scoring is associated with extra efforts regarding time and labor and requires veterinarian expertise, but heritabilities were larger compared with a binary DD definition based on producer records (Schöpke et al., 2015).Consequently, in genomic analyses, stronger association signals for molecular markers and annotated potential candidate genes are expected when considering a detailed phenotyping strategy reflecting disease pathogenesis.In genome-wide association studies (GWAS) for simple binary DD with subsequent gene annotations, potential candidate genes were detected on chromosomes 3, 6, 9, 11, 12, 19 and 26 (Naderi et al., 2018;Kopke, 2019;Sanchez-Molano et al., 2019;Shabalina et al., 2020b).However, most of the significantly associated SNPs only surpassed the less conservative suggestive threshold, but were not significant according to the strict Bonferroni criterion.In addition to precise phenotyping, detailed modeling of environmental effects might contribute to a deeper understanding of genomic mechanisms.For disease indicator traits, obvious genotype x environment interactions were detected when modeling the environment through herd gradients reflecting herd size or the hygiene status (Yin and König, 2018).Environmental alterations might contribute to differences in gene expressions, as experimentally shown in heat stressed mice (Cammack et al., 2009).In dairy cattle genomic studies, Shabalina et al. (2021) identified different significant SNPs for DD in GWAS conducted either in organic or in conventional populations.Sölzer et al. (2022) focused on genotype by climate interactions, and annotated different potential candidate genes for DD under heat stress and thermoneutral conditions.In addition to population or climatic effects, specific housing characteristics might contribute to the claw health status and possible genotype x housing interactions for DD.In the context of DD improvements, Lobeck et al. (2011) promoted compost bedded pack barns (CBPB) as a favorable housing alternative.In conventional free stall herds (CON), the phenotypic expression of lameness is mostly induced through the concrete floor (Somers et al., 2005;Kester et al., 2014).However, different breeds responded differently in CBPB and CON, pointing to possible genotype x housing system interactions (Leso et al., 2020).In this regard, Wagner et al. (2021) detected specific significant SNPs for udder health due to housing particularities.The annotated potential candidate genes were related to the regulation of the immune system under environmental stress.
Consequently, the aim of the present study was to conduct comprehensive genetic and genomic analyses for novel traits derived from DD scoring including (1) the estimation of genetic parameters in a single-step approach separately for the overall, the CON and the CBPB data sets, and (2) GWAS and the annotation of potential candidate genes for main and interaction effects with the discrete housing system considering the phenotypic records from genotyped cows.

Digital dermatitis recording
Phenotyping for DD traits considered 2,980 observations of 1,311 lactating Holstein-Friesian (HF) and 399 Fleckvieh-Simmental (FS) cows.The DD scoring in the years 2021 and 2022 was accomplished according to a validated diagnosis scheme (Döpfer et al., 1997), considering the DD stages M.0 to M.4.1 of the claws from the hind legs.For DD stage diagnosing, the cows were scored by one trained veterinarian.In this regard, the veterinarian has seen all cows from the herd at the same recording date.The interval of consecutive herd visits comprised 3 mo.A specific mirror was used to diagnose the different DD stages as accurate as possible.The  Döpfer et al. (1997) and the stage transformations into the above mentioned DD traits.For different observed DD stages at different cow legs, the respective cow was considered as sick = 1 for, e.g., a DD-chronic entry at the right hind leg, and as sick = 1 for, e.g., a DD-acute entry at the left hind leg.Hence, all DD stages were analyzed separately considering the binary data structure (no entry for the respective DD stage implied a score = 0 for healthy), and using the cow as the observational unit.The average prevalence (always calculated in relation to the healthy control unit from the same herd visit) was 20.47% for DD-sick, 13.88% for DD-acute and 5.34% for DD-chronic.

Housing system characteristics
The cows were kept in 15 farms located in the German federal states of Bavaria, Hesse, Rhineland-Palatinate and Brandenburg reflecting the housing system CON or CBPB.The CON farms were typical German cubicle farms with concrete floor.The herd size per farm ranged from 25 to 840 cows.Lactation numbers of cows ranged from one to 14. Five farms only had the CON system, 5 other farms only the CBPB system, and the remaining 5 farms both systems CON and CBPB.The farms with both systems had 2 farm buildings, i.e., one for the CBPB cows and the other one for the CON cows, which were considered as to 2 separate herds in the ongoing analyses.The CON herds were chosen as control herds for a respective CBPB herd with regard to geographic coordinates, climate, feed ration and feeding system, and herd size according to the guidelines for selecting case (i.e., compost) and control (i.e., cubicle) herds as defined by an expert group from the European Freewalk Consortium (Blanco-Penedo et al., 2020).The prevalences were 10.65% and 26.39% for DD-sick, 7.45% and 17.11% for DD-acute, and 1.49% and 9.03% for DD-chronic, for cows kept in CBPB and CON, respectively.

Genotypes
A subset of 935 cows was genotyped with the Illumina BovineSNP50 v2 BeadChip (Illumina Inc.).Quality control of the genotype data was performed using the software package PLINK (Purcell et al., 2007).SNPs with a minor allele frequency <1% and a significant deviation from the Hardy-Weinberg equilibrium (P < 1 × 10 −8 ) were discarded.Finally, 38,495 SNPs from 926 cows remained for the genomic studies.The average coefficient for relationships based on SNP data between the HF and FS cows from Bavaria was 4.24%, indicating the utilization of Red Holstein bulls in the Simmental population in past decades.Numbers of cows and observations with phenotypes and genotypes in both production systems CON and CBPB are given in Table 1.

Single-step genetic parameter estimations.
For the estimation of genetic (co)variance components and breeding values for DD-sick, DD-chronic and DDacute via ssGBLUP, the genetic-statistical single-trait repeatability model 1 was defined in matrix notation as follows: where y was a vector including observations for DDsick, DD-chronic or DD-acute; β was a vector of fixed effects including herd, year of diagnosis, season of diagnosis (4 seasons: Jan -Mar.; Apr.-Jun.;Jul.-Sep.; Oct.-Dec.),parity (5 classes: 1, 2, 3, 4, 5 and >5) and breed; a was a vector of random additive genetic effects, with a ~N(0, H σ 2 a ) where σ 2 a was the additive genetic variance; pe was the vector of random permanent environmental effects for repeated measurements across lactations, with pe ~N(0, Iσ 2 pe ) where σ 2 pe was the permanent environmental variance; e was the vector of random residual effects with e ~N(0, Iσ 2 e ) where σ 2 e was the residual variance; X, Z and W were the incidence matrices for fixed, additive genetic and permanent environmental effects, respectively.The combined inverse of the H matrix was computed according to Legarra et al. (2009) by considering G w = (0.95 x G + 0.05 x A 22 ), where A 22 was the submatrix of the pedigree-based relationship matrix for genotyped animals and G was the genomic relationship matrix (VanRaden, 2008).The pedigree relationship matrix considered ancestors back to at least 3 generations and comprised 11,385 animals with genetic relationships to the cows with phenotypes.The oldest ancestor in the pedigree data set was born in 1921.Model 1 was applied to the whole data set, as well as for the sub-data sets CON and CBPB.
Genetic correlations between DD-sick CBPB with DDsick con , between DD-chronic CBPB with DD-chronic CON , and between DD-acute CBPB with DD-acute CON were estimated via bivariate animal models, and considering the same DD trait as 2 different traits.The effects were the same as defined for model 1.However, a cow was kept in the same housing system during the recording period, implying not estimable residual covariances and residual correlations.Genetic correlation estimates were compared with breeding value correlations (breeding values from the single-trait analyses) considering phenotyped cows and sires with DD daughter records.
Genetic correlations between DD-sick, DD-chronic and DD-acute for the whole data set considering cows from both housing systems were estimated in bivariate repeatability models, considering the same fixed and random effects as described above for model 1.Nevertheless, due to the DD trait definitions, DD-chronic and DD-acute represent a sub-sample of DD-sick, implying some kind of auto-correlations due to trait dependencies.
All (co)variance components were estimated using the AI-REML algorithm as implemented in the BLUPF90 software package (Misztal et al., 2018).
Genome-wide associations for DD stages with housing system interactions.GWAS only considering the phenotypic records from genotyped cows (see Table 1) were applied to estimate SNP main effects and SNP x housing system interaction effects for DDsick, DD-acute and DD-chronic.For the interaction analyses, we only considered one observation per cow and DD trait, i.e., the DD record from the most recent lactation.Calculations were performed using our own software package GWAInter.R (Halli et al., 2021), and applying generalized least squares methodology and the algorithm as introduced by Yang et al. (2014).The respective statistical model 2 to estimate main and interaction SNP effects was: Trait definition as explained in the materials and methods and in Figure 1.
where y = a vector of observations for DD-sick, DDchronic or DD-acute ; x snpi = a vector for SNP genotypes; u snpi = a vector including SNP main effect; x interi = a vector of genotypes for cows in the compost bedded pack barn system; u interi = a vector for the difference in regression coefficients, i.e., SNP effects in the compost bedded pack barn system compared with SNP effects in the cubicle housing system (SNP by housing system interaction effect).The notations for fixed and random effects are described in model 1.The modeling approach implied 2 P-values for each SNP, i.e., one for the significance of the main effect and another one for the significance of the interaction effect.The genome-wide significance level according to Bonferroni was defined with pBF = 0.05/ no. of SNP = 1.3e-06.A normative significance threshold was used to identify potential candidate SNP, considering pCD = 1e-04 (Kurz et al., 2019).Annotated potential candidate genes located in 100 kb upstream and downstream from the significantly associated candidate SNP were detected using Ensembl, release 102 (Zerbino et al., 2018).

Genetic parameters for DD traits
The estimates for single-step variance components and heritabilities are given in Table 2.The heritabilities for the 3 DD traits were in a narrow range with 0.16 for DD-sick, 0.14 for DD-acute and 0.11 for DDchronic.The more precise phenotyping of DD based on veterinarian expertise contributed to larger genetic variances and heritabilities compared with producer health diagnoses as outlined by Schöpke et al. (2015) for pedigree based genetic relationships.This is in agreement with results for other health or functional traits.For udder health trait genetic analyses, Wagner et al. (2021) emphasized the more precise genetic evaluations when using laboratory measurements including specific mastitis pathogens compared with analyses based on producer recorded clinical mastitis.With regard to female fertility and classical trait definitions as used in genetic evaluations, Berry et al. (2014) reported small heritabilities ranging from 0.01 to 0.18.Heritabilities were substantially larger up to 0.33 with genetic studies on the underlying endocrine profiles such as measurements for progesterone (Häggman et al., 2019).However, a more accurate phenotyping strategy is always associated with additional efforts regarding logistics, time and labor, hampering the creation of very large data sets.
In addition to the phenotyping strategy or the trait definition, environmental conditions might contribute to genetic parameter differences between studies.In the present study, heritabilities and variance components for same DD traits were quite similar for the overall data set, CBPB and CON (Table 2).Nevertheless, additive genetic, permanent environmental and residual variances for all DD traits were larger in the CON environment.With regard to the comparison of same variance components between the systems CON and CBPB for the same DD trait, the larger differences for additive genetic variances than for the permanent environmental and residual components contributed to the smallest DD heritabilities in the CBPB sub-data set.The CBPB production system reflects a modern "well-being environment," with strong focus on cow welfare (Leso et al., 2020).Schierenbeck at al. (2011) reported larger standard deviations for estimated breeding values and daughter yield deviations, indicating a more pronounced genetic differentiation, in "challenging environments" for low heritability health traits.Such finding was supported by Schafberg et al. (2006) for detailed recorded mastitis traits including specific major pathogens.In such context, the lower prevalence of diseases in superior environments, i.e., a higher fraction of healthy cows in CBPB than in the more challenging CON system, might hamper the iden- The genetic correlations for same DD traits from different housing systems were 0.85 between DD-sick CBPB with DD-sick CON , 0.73 between DD-acute CBPB with DD-acute CON , and 0.82 between DD-chronic CBPB with DD-chronic CON (Table 3).The genetic correlation estimates had quite large SE, but the respective breeding value correlations (based on the breeding values from the single-trait analyses) were very similar.The breeding value correlations (Pearson correlation coefficient) between same DD traits from the different data sets (whole data set, CBPB, CON) for cows with phenotypic records as well as for 61 sires with more than 5 phenotyped daughters in both systems were in the range from 0.68 (DD-acute CON with DD-acute CBPB for cows) to 0.83 (DD-sick CON with DD-sick whole data set for sires).Only in case of a breeding value accuracy of 1, a correlation between breeding values is identical with a genetic correlation (Calo et al., 1973).According to the theoretical derivations made by Robertson (1959), correlations smaller than 0.80 indicate significant genotype x housing interactions.The genetic correlations and breeding value correlations close to 0.80 (see Table 3) indicate only slight re-rankings of animals in the different housing systems.Substantially smaller genetic correlations for health traits were reported by Shabalina et al. (2021) when creating data sets according to organic or conventional criteria.Such classification simultaneously comprises feeding, housing and management aspects.The CON -CBPB classification strongly focused on the housing separation, while keeping feeding, milking and management effects as identical as possible.For different data sets of largescale CON herds reflecting different feeding systems, Yin and König (2018) estimated very small genetic correlations between same health and health indicator traits.
Considering the overall data set including cows from both housing systems, the genetic correlations between the different DD traits were 0.81 (DD-sick with DD-acute), 0.93 (DD-sick with DD-chronic) and 0.58 (DD-acute with DD-chronic) (Table 4).Hence, consideration of specific DD stages and disease pathogenesis according to veterinary expertise (Döpfer et al., 1997) reflect a differing genetic background for DD-acute and DD-chronic.The genetic correlations between DD-sick with DD-chronic, and between DD-sick with DD-acute, might be biased due to the existing auto-correlations and trait dependencies (DD-acute and DD-chronic are subsets of DD-sick).The phenotypic correlations among the DD traits were slightly smaller than the respective genetic correlations (Table 4).

Genome-wide associations and potential candidate genes for DD traits
The Manhattan plot from the GWAS for SNP main effects of the 3 DD-traits is shown in Figure 1, with DD-sick in Figure 1a.The corresponding inflation factor λ was 1.32.We identified 23 significant SNPs for DD-sick associated with 24 potential candidate genes as summarized in the supplementary material (supplemental table S1).However, only one of these SNPs (on BTA 29) exceeded pBF.This SNP is located in the exon of the gene TENM4.Generally, the largest number of significant SNPs according to the more relaxed normative threshold was detected on BTA 29.The annotated potential candidate gene UMODL1 on BTA 1 is related to immune response and regulations of neutrophil granulocyte migration (Fortin et al., 2019).Neutrophil granulocytes are the category of defense cells responding first to bacterial infections, such as in case of DD pathogens.In humans, UMODL1 regulated immune and reproductive mechanisms (Wang et al., 2012), and recently has been identified as a candidate gene for female fertility in Brown Swiss cattle (Manca et al., 2021).The annotated potential candidate gene NUGGC on BTA 8 is involved in the cellular response to alterations of lipopolysaccharide levels, e.g., in the  Trait definition as explained in the materials and methods and in Figure 1.
formation of the outer cell membrane of gram-negative bacteria (Raetz and Whitfield, 2002).Linking NUGGC to DD, gram-negative treponemes represent the main bacterial milieu of a DD infection (Döpfer et al., 2012).Furthermore, NUGGC was intensively discussed in the context of foot-and-mouth disease infections, with overexpression during humoral immune response (Eschbaumer et al., 2016).The 2 potential candidate genes CCDC88C and PPP4R3A on BTA 21 were associated with metabolic cow diseases in early lactation (Klein et al., 2021).Metabolic diseases contribute to stress and impair the immune defense mechanisms, in causality implying an increased susceptibility to claw infections (Schöpke et al., 2015).The Manhattan Plot from the GWAS for DD-acute GWAS is presented in Figure 1b.The corresponding inflation factor λ was 1.31.We identified 21 significant SNPs associated with 24 potential candidate genes as summarized in the supplementary material (supplemental table S1).One of these SNPs on BTA 23 exceeded pBF.The largest number of significant SNPs according to the normative threshold was detected on BTA 6 and BAT 11.The potential candidate gene NRXN3 on BTA 10 affected neuronal receptors, specifically the differentiation of synapses (Reissner et al., 2013).Functions of NRXN3 mainly addressed behavioral disorders and temperament abnormalities in mice (Brown et al., 2011) and in Brahman cattle (Paredes-Sanchez et al., 2020).In dairy cattle genomic health studies, Klein et al. (2020) and Hayirli ( 2006) associated NRXN3 with metabolic disorders.The potential candidate gene UBR5 on BTA 14 is involved in the ubiquitination of proteins, but was deregulated in case of severe disease stressors, especially due to some cancers (Shearer et al., 2015).Van der Spek et al. (2015) related genotypes of UBR5 to the claw disease interdigital hyperplasia in Holstein cows.Strong genetic correlations between interdigital hyperplasia and DD were estimated in pedigree-based analyses (König et al., 2005;Gernand et al., 2012), supporting the results from genomic studies and the identified overlapping genomic mechanisms (Sölzer et al., 2022).The annotated potential candidate gene DOCK2 on BTA 20 for DD-acute is directly involved in immune defense mechanisms.A mutation in this gene implied immunodeficiency in humans, due to impaired functions of T cells, and defects of B and NK cells (Dobbs et al., 2015).
The Manhattan Plot for the SNP main effects of DD-chronic GWAS is presented in Figure 1c.The corresponding inflation factor λ was 1.39.We identified 25 significant SNPs associated with 33 potential candidate genes as summarized in the supplementary material (supplemental table S1).Two of these SNPs exceeded pBF, one located on BTA 11 and the other one located on BTA 16.The largest number of significant SNPs according to pCD was detected on BTA 11 and 28.The significant SNP ARS-BRGL-NGS-23211 on BTA 11 is located in a segment of the gene WDR54, as well as in close chromosomal distance to 16 further potential candidate genes.All of these genes, i.e., AUP1, TLX2, PCGF1, LBX2, CCDC142, MRPL53, WBP1, RTKN, WDR54, C11H2orf81, MGC152281, ENSBTAG000050345 and DCTN1, were associated with multifocal paratuberculosis in cattle (Canive et al., 2021).Paratuberculosis is caused by Mycobacterium avium ssp.paratuberculosis, and very similar to DD, the phenotypic expression of this multifactorial disease is strongly related to immune system deficiencies (Canive et al., 2021).The annotated potential candidate gene TACR2 on BTA 28 is involved in inflammatory processes and immune defense mechanisms, and was described in the context of substance P release in pigs (Jakimiuk et al., 2017).Substance P causes vasodilation and an increase in vascular permeability, and is responsible for the chemotaxis of leukocytes (Harrison and Geppetti, 2001).Some overlapping association signals were found when comparing the Manhattan plots for DD-sick and DD-acute, and for DD-sick and DD-chronic.For DD-sick and DD-acute, the annotated potential candidate gene METTL25 on BTA 5 is involved in DNA methylation (de Greef et al., 2023).Another shared potential candidate gene, i.e., AFF3 on BTA 11, was discussed in the context of expressions in B cells and oncogenesis (Shi et al., 2018), and associated with abortion in Holstein heifers (Oliver et al., 2019).The shared potential candidate gene PRKG1 on BTA 26 contributes to the cGMP signaling pathway (Hofmann et al., 2009), and was associated with milk fatty acid contents in Chinese Holstein populations (Shi et al., 2019).A further shared potential candidate gene for DD-sick and DD-acute is TENM4 on BTA 29.TENM4 is a marker for myoblast quiescence (Pietrosemoli et al., 2017), and was discussed in the context of zilpaterol application in cattle (Reith et al., 2022).Zilpaterol is a β agonist with effects on muscle mass accumulation in farm animals.However, zilpaterol application under Trait definition as explained in the materials and methods and in Figure 1.
heat stress conditions contributed to a downregulation of TENM4 expression (Reith et al., 2022), indicating an environmental effect on gene activities.
A shared potential candidate gene for DD-sick and DD-chronic is MCU on BTA 28.The protein encoded by MCU interacts with mitochondrial calcium uptake (Tarasov et al., 2012).Calcium in turn plays an important role in milk production (Breves et al., 2016), as well as in keratinization and mature horn cell formation of the claw (Langova et al., 2020).Rodríguez et al. (2017) associated increased blood calcium losses at the beginning of lactation and hypocalcemia of high-yielding cows with several diseases including claw disorders.The potential candidate gene ITGA11 for DD-sick and DD-chronic on BTA 10 is involved in integrin-mediated cell adhesion and cell migration (Zhang et al., 2002), and was associated with DD in beef cattle (Kopke et al., 2020).

Genome-wide associations for SNP interaction effects with housing system effects
The Manhattan plots for the SNP interaction effects with the housing system are displayed in Figure 2 (DD-sick), Figure 3 (DD-acute) and Figure 4 (DDchronic).As indicated above, a significant interaction means that the respective SNP is relevant for a housing system CBPB, but not for CON, or vice versa.The corresponding inflation factor λ was 1.17 for DD-sick, 1.2 for DD-acute and 1.47 for DD-chronic.The annotated potential candidate genes for interaction effects are listed in Table 5.For DD-sick and DD-acute, 3 and 5 SNPs, respectively, exceeded the candidate threshold, but no SNP was significantly associated with DDchronic.For DD-sick and DD-acute, the same SNP on BTA 13 was significant for the interaction component.This SNP is located in close distance to the genes ASXL1 and NOL4L.Both genes were associated with the occurrence of cancer (Stein et al., 2011;Lin et al., 2021).Naderi et al. (2020) reported significant effects of ASXL1 and NOL4L on binary DD when evaluating high-yielding HF cows from CON systems.However, they could not confirm their findings in a sub-sample of black and white dual-purpose cattle (founder breed of modern HF), which were mostly kept in grassland systems.Hence, effects of the breed and / or of the system contributed to SNP significances.Additionally, NOL4L was suggested as potential candidate gene for udder health (Wolf et al., 2021).More than 95% of all somatic cells are leucocytes, and levels of leucocytes have been suggested as indicator for bacterial infections of the udder (Bradley and Green, 2005).Consequently, Gernand et al. (2012) postulated a close genetic relationship between somatic cell score and bacterial claw infections including DD, but the genetic correlation between DD and SCS was close to zero.For DD-acute, 2 significant SNPs for the interaction component are located on BTA 2 in close distance to the genes SCN2A and SCN3A.Both genes encode proteins that are involved in the construction of voltage-gated sodium channels, playing an important role in the transmission of action potentials in neurons.Mutations in the SCN2A and SCN3A genes were associated with epilepsy in humans (Holland et al., 2008).Another significant SNP for the interaction effect of DD-acute on BTA 4 is located in close chromosomal distance to the genes LEP, RBM24 and IMPDH1.LEP is involved in many cellular signal pathways in the organism, including the regulation of the innate and adaptive immune system.For example, LEP promotes the activation of neutrophil granulocytes (Francisco et al., 2018).Neutrophil granulocytes, in turn, are involved in initial defense mechanisms against bacteria through phagocytosis and exocytosis (Guo and Ward, 2005), explaining the effects on DD.In dairy cattle, Shabalina et al. (2020a) reported significant effects of LEP and of RBM28 on the length of productive life for cows kept in conventional dairy farms, but they could not verify their results in organic environments (Shabalina et al., 2021).The annotated potential candidate gene IMPDH1 was involved in immunological signal pathways (Malvisi et al., 2016;Jonsson and Carlsten, 2002).In heat-stressed Angus cattle, the expression of IMPDH1 was downregulated under simultaneous zilpaterol application, as previously indicated for TENM4 (Reith et al., 2022).The significant SNP for the interaction component of DD-acute is located on BTA 7. One annotated potential candidate gene is MRPL55, which is involved in protein biosynthesis in the mitochondrion and encodes a protein of the 39 S-subunit (Mai et al., 2017).MRPL55 was associated with early abortions in mice (Cheong et al., 2020).In Brown Swiss cattle, MRPL55 was suggested as a potential candidate gene for female fertility with largest effects on the interval between first and last insemination (Häfliger et al., 2021).The identified potential candidate gene for the DD-acute interaction effect is located on BTA 7, and was associated with milk yield traits in Chinese Holstein cows (Lu et al., 2022).

CONCLUSION
Heritabilities for the 3 DD traits considering precise phenotyping based on a veterinarian expertise were slightly larger compared with previous estimates for DD-producer records.Heritabilities and additive genetic variances were slightly larger for same DD traits in the CON than in the "well-being" CBPB system, indicating a stronger genetic differentiation of diseases in a more challenging conventional environment.However, genetic correlations between same DD traits in the different systems CON and CBPB were close to 0.80, disproving possible genotype x housing interactions.The genetic correlation between DD-acute and DD-chronic was 0.58, indicating a partly different genetic background for acute and chronic diseases.The GWAS for main SNP effects indicated heterogeneous Manhattan plots especially for DD-acute and DDchronic, supporting the differences or particularities in disease pathogenesis.Nevertheless, we identified a few shared annotated potential candidate genes, i.e., METTL25, AFF3, PRKG1 and TENM4 for DD-sick and DD-acute, which mostly have been reported in the context of immunology.For SNP x housing system interactions, only a few significant SNPs were detected.From a practical perspective, the only small genotype x housing interaction effects support the current genetic evaluation practice, i.e., there is no need to consider specific herd husbandry or herd management characteristics.Trait definition as explained in the materials and methods and in Figure 1. 2 No gene name or no gene position: SNP was not located in the gene or within a window size of 100 kb up-and downstream.
Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS single DD stages were grouped into 3 traits: DD-sick (including the stages M.1-M.4.1),DD-acute (including the stages M.1, M.2 or M.3) and DD-chronic (including the stages M.4 or M.4.1).The stage M.0 represents the healthy cows without any lesions.Figure 1 illustrates the different DD stages according to Figure 1.Digital dermatitis (DD) stages according to the diagnosis scheme by Döpfer et al. (1997) and classification into the DD traits DD-sick, DD-acute and DD-chronic.
Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS Table 4. Genetic (above diagonal) and phenotypic correlations (below diagonal) among the different DD-traits (analyses based on the overall data set including cows from both housing systems) Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS Figure 2. a) Manhattan plot for -log (10) P-values of SNP main effects for DD-sick, b) DD-acute, c) DD-chronic.The DD traits are explained in the materials and methods and in Figure 1.

Figure 3 .
Figure 3. Manhattan plot for -log (10) P-values of SNP interaction effects with the housing system for DD-sick.The DD traits are explained in the materials and methods and in Figure 1.

Figure 4 .
Figure 4. Manhattan plot for -log (10) P-values of SNP interaction effects with the housing system for DD-acute.The DD traits are explained in the materials and methods and in Figure 1.

Figure 5 .
Figure 5. Manhattan plot for -log (10) P-values of SNP interaction effects with the housing system for DD-chronic.The DD traits are explained in the materials and methods and in Figure 1.

Table 1 .
Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS Number of observations, number of cows and prevalence for the three DD traits 1 in the housing systems compost bedded back barn (CBPB) and conventional cubicle (CON)

Table 2 .
Sölzer et al.: GENOMIC x HOUSING INTERACTIONS FOR DIGITAL DERMATITIS Heritabilities (h 2 ) with corresponding SE in brackets, additive genetic variances (σ 2 of genetic differences.Effects of incidences on genetic variations and heritability estimates have been theoretically derived for threshold model applications and human diseases by Dahlquist et al. (2019).Consequently, using a health data set from co-operator herds reflecting a superior environment, Gernand et al. (2013) only focused on major claw disorders displaying lactation incidences larger than 10%, due to failure in convergence or extremely small genetic variances for rare diseases. tification

Table 3 .
Genetic correlations (SE in brackets) and breeding value correlations (from the cows with own digitalis dermatitis (DD) records) between same DD traits 1 recorded in the different environments compost bedded pack barn (CBPB) and conventional barns (CON)

Table 5 .
Significantly associated SNP markers and annotated potential candidate genes for the interaction effect with the housing system