Associations between serum health biomarker concentrations and reproductive performance, accounting for milk yield, in pasture-based Holstein cows in south-eastern Australia

In this single cohort study, we investigated associations between the concentrations of a suite of serum biomarkers measured in the first 30 d of lactation and subsequent reproductive performance measured as mating start date to conception intervals, in pasture-based Holstein cows. A secondary objective was to examine associations between biomarker concentrations and 305-d milk yield to assess whether any positive associations between biomarker concentration and reproductive performance were explained by reduced milk production. The data used had been collected as part of an ongoing project from 2017 to 2020 to compile a data set from a large population of lactating dairy cows. Biomarkers measured were those associated with energy balance (β-hydroxy butyrate (BHB) and nonesterified fatty acids (NEFA)), protein nutritional status (urea and albumin), immune status (globulin, albumin to globulin ratio and haptoglobin) and macro-mineral status (calcium and magnesium). Associations between biomarker concentrations and mating start date to conception interval were investigated using Cox proportional hazards models using between 634 and 1,121 lactations (varying by biomarker) from 632 to 1,103 cows and 11 to 17 mating periods from 10 to 13 herds. Based on hazard ratio (HR) estimates and associated 95% confidence intervals, hazard of conception on any particular day of the herds’ mating periods was positively associated with the concentrations of albumin (HR = 1.09; 95% CI: 1.05 to 1.12), albumin to globulin ratio (HR = 2.82; 95% CI: 1.66 to 4.79), calcium (HR = 2.01; 95% CI: 1.18 to 3.43) and magnesium (HR = 2.17; 95% CI: 1.01 to 4.66),


ABSTRACT
In this single cohort study, we investigated associations between the concentrations of a suite of serum biomarkers measured in the first 30 d of lactation and subsequent reproductive performance measured as mating start date to conception intervals, in pasturebased Holstein cows.A secondary objective was to examine associations between biomarker concentrations and 305-d milk yield to assess whether any positive associations between biomarker concentration and reproductive performance were explained by reduced milk production.The data used had been collected as part of an ongoing project from 2017 to 2020 to compile a data set from a large population of lactating dairy cows.Biomarkers measured were those associated with energy balance (β-hydroxy butyrate (BHB) and nonesterified fatty acids (NEFA)), protein nutritional status (urea and albumin), immune status (globulin, albumin to globulin ratio and haptoglobin) and macromineral status (calcium and magnesium).Associations between biomarker concentrations and mating start date to conception interval were investigated using Cox proportional hazards models using between 634 and 1,121 lactations (varying by biomarker) from 632 to 1,103 cows and 11 to 17 mating periods from 10 to 13 herds.Based on hazard ratio (HR) estimates and associated 95% confidence intervals, hazard of conception on any particular day of the herds' mating periods was positively associated with the concentrations of albumin (HR = 1.09; 95% CI: 1.05 to 1.12), albumin to globulin ratio (HR = 2.82; 95% CI: 1.66 to 4.79), calcium (HR = 2.01; 95% CI: 1.18 to 3.43) and magnesium (HR = 2.17; 95% CI: 1.01 to 4.66), and negatively associated with globulin concentration (HR = 0.98; 95% CI: 0.97 to 1.00).There was also some evidence that NEFA concentration was negatively associated (HR = 0.76; 95% CI: 0.57 to 1.01), and urea concentration positively associated (HR = 1.05; 95% CI: 0.99 to 1.11), with reproductive performance, but no evidence that BHB and haptoglobin concentrations were associated with reproductive performance.Except for NEFA, presence and direction of the association between the biomarker and milk yield was not discordant with that for reproductive performance.Also, except for NEFA, we found no substantial evidence of nonlinear relationships between biomarker concentration and either reproductive performance or milk yield.Correlations between biomarker concentrations were generally weak, indicating that multi-biomarker panels may collectively predict reproductive performance better than any single biomarker.We noted substantial variation in the concentrations of all biomarkers within, and for

INTRODUCTION
Health disorders in early lactation have negative effects on both milk production and subsequent reproductive performance in dairy cows (Suthar et al., 2013;McArt et al., 2015).The magnitude of these negative effects can be estimated using observational studies that investigate associations between objectively measured health indicators and subsequent reproductive and production outcomes.A common methodology for Associations between serum health biomarker concentrations and reproductive performance, accounting for milk yield, in pasture-based Holstein cows in south-eastern Australia objectively determining the health and nutritional status of cows is serum metabolic profiling, which involves measuring the concentration of a suite of biomarkers in serum (Payne et al., 1970, Macrae et al., 2006).Commonly used biomarkers in metabolic profiling include those associated with energy balance; β-hydroxy butyrate (BHB) and nonesterified fatty acids (NEFA), protein nutritional status; urea and albumin, immune status; globulin, albumin to globulin ratio and haptoglobin, and macro-mineral status; calcium and magnesium (Whitaker, 2004, Anderson, 2009).
Most studies to investigate associations between serum biomarker concentrations and reproductive performance in dairy cattle have assessed the commonly used biomarkers of energy balance, BHB and NEFA.Perturbations of energy balance have long been associated with adverse health and production outcomes, and results from these studies consistently show that elevated concentrations of both biomarkers are associated with reduced reproductive performance (Ospina et al., 2010a, McArt et al., 2013, Compton et al., 2014).Serum urea concentration is used as a biomarker of rumen degradable protein intake (Macrae et al., 2006) and protein intake and protein nutritional status (Roseler et al., 1993), and several studies have demonstrated that elevated serum urea concentrations are associated with lower pregnancy and conception rates (Butler et al., 1996, Raboisson et al., 2017).Relatively few studies have reported associations between serum protein profiles (concentrations of albumin and globulin, and the ratio of albumin to globulin) and reproductive performance (Rowlands et al., 1977, Rowlands and Manston, 1983, Cattaneo et al., 2021).Reported associations between subclinical macro-mineral disorders such as hypocalcemia and reproductive performance in grazing systems are also not consistent.For example, Umaña Sedó et al. (2018) reported that cows experiencing subclinical hypocalcemia on the day of calving had a lower hazard of pregnancy and took 32 d longer to become pregnant compared with normocalcemic cows.In contrast, Roberts and McDougall (2018) reported that subclinical hypocalcemia was not associated with any of the reproductive outcomes measured.Similarly, the effect of periparturient hypocalcemia on subsequent milk production is reported to vary depending on both the time and duration of the hypocalcemic event, and parity (Neves et al. 2018a,b, McArt andNeves, 2019).
Most of the aforementioned epidemiological studies examined associations between only one or 2 biomarkers and reproductive performance.There is, however, increasing interest in the use of multi-biomarker panels to better define and describe the health and nutritional status of cows, such as a metabolic clustering approach described by multiple authors (Tremblay et al., 2018, DeKoster et al., 2019, Xu et al., 2019).But most of these approaches remain focused on a single category of biomarker, such as indicators of energy balance.Furthermore, to date there have been no large-scale observational studies which have investigated associations between such biomarker panels and reproductive performance in dairy cows.Given the complex interrelationships between metabolic, infectious, and inflammatory disorders during early lactation, there is value in modeling the associations between a suite of biomarkers that constitute a more holistic metabolic profile and reproductive performance.This complexity may also be better understood by simultaneously modeling associations between multiple biomarkers and reproductive performance.
The aim of many observational studies investigating associations between biomarkers and health, reproduction and production outcomes has been to define thresholds or cut-off values, above or below which adverse events are more likely to occur.Receiver operating characteristic (ROC) analysis has been commonly used to determine the biomarker threshold at which maximum combined diagnostic sensitivity and specificity is achieved (Ospina et al., 2010b, Compton et al., 2014).This approach has the advantage of simplifying the interpretation and communication of results, which is important when results are being used for individual animal diagnostic purposes.However, by defining a threshold, one is implicitly assuming that no important relationship exists within the range of values below the threshold, and also that no important relationship exists within the range of values above that threshold.This inherently risks losing potentially valuable information that is available when biomarkers are interpreted as continuous data.This is particularly important if associations are used in economic studies to quantify the impact of a management or nutritional intervention, where a dose-response curve exists and even a numerically small change in response is potentially important.For example, even if a change in ration results in modest changes in a biomarker concentration which, in turn, is linearly associated with a small change in reproductive performance, the economic benefits of that ration change could still be substantial.Accordingly, it is important to assess relationships between biomarkers and reproductive performance for non-linearity.
The vast majority of biomarker epidemiological studies in dairy cattle have been performed in cows fed a total mixed ration (TMR), and the applicability of findings from these studies in pasture-based systems, which are commonplace in south-eastern Australia, is unclear (Raboisson et al., 2017).High herd reproductive performance is an important driver of productivity and profit in Australian pasture-based dairy systems.

Luke et al.: Associations between health biomarkers and fertility
In addition to allowing voluntary culling and optimizing herd replacement rate, high reproductive performance ensures that cows calve at optimal times during the year.In pasture-based dairy systems, farm profit is closely related to the amount of home-grown fodder consumed by cows (Chapman et al., 2014), and high herd reproductive performance allows farmers to align herd calvings with maximum pasture availability.While a few studies have investigated associations between serum biomarker concentrations and reproductive performance in pasture-based systems in New Zealand (Compton et al., 2014, Roberts andMcDougall, 2019), there is a dearth of studies from Australia where, by comparison, (1) production systems are far more diverse (Dairy Australia, 2017), (2) concentrate feeding makes up a higher proportion of dry matter intake, and (3) per cow production is substantially higher (Dairy Australia, 2019, DairyNZ, 2021).
The primary objective of this study was to investigate associations between the concentrations of a suite of serum health biomarkers measured in the first 30 d of lactation, and the subsequent reproductive performance of lactating dairy cows from pasture-based farms in south-eastern Australia, including assessing relationships for linearity and curvilinearity.A secondary objective was to investigate associations between biomarker concentrations and milk yield.Our reason for doing this was to examine whether any positive associations between biomarker concentration and reproductive performance were explained by reduced milk production .Our a priori hypotheses were that 1) the concentrations of one or more of these biomarkers are associated with reproductive performance or milk yield, 2) that these associations are nonlinear, 3) that these associations are influenced by the calving to sampling interval, and 4) that associations with reproductive performance are discordant with associations with milk yield.Our goal was to provide a broad overview of the epidemiology of biomarkers commonly employed in an holistic metabolic profile, and to establish dose-response curves between biomarker concentrations and outcomes of interest which can be used in future economic and genetic analyses.

MATERIALS AND METHODS
All procedures undertaken in this study were conducted in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (National Health and Medical Research Council, 2013) 2017-07, 2018-07, 2018-08, 2019-09 and 2019-11.

Study Population and Study Design
A single cohort study of dairy cows in south-eastern Australia was conducted using previously collected data.The data set had been collected as part of an ongoing project compiling a data set from a large population of lactating dairy cows for development of 1) milk Fourier-transform mid-infrared spectroscopic predictions of metabolic disorders (Luke et al., 2019b, Ho et al., 2021), and 2) genomic breeding values for improved transition cow health (Luke et al., 2019a).Serum metabolic profiles were available from 5,211 lactations from 4,199 cows from 24 herds in south-eastern Australia.
Blood samples were collected from August 2017 to March 2020 using the protocol described in Luke et al. (2019b).Briefly, study herds were visited regularly during the herd calving periods and clinically healthy cows that had calved within the last 30 d were sampled.Thus, cows were sampled at varying intervals after calving.Ten ml blood samples were collected from the coccygeal vein of cows into 10 mL serum clot activator vacutainer tubes (Becton Dickinson, Franklin Lakes, NJ, USA), at or immediately after milking.Sera were analyzed for all samples: BHB, NEFA, and urea, and in subsets of samples: albumin, total protein, haptoglobin, calcium, and magnesium, using a Kone 20 XT clinical chemistry analyzer (Thermo Fisher Scientific, Waltham, MA).Details of assays and reagents are described in Luke et al., 2019a.Globulin concentrations were calculated as total protein concentration minus albumin concentration, and albumin to globulin ratio as albumin concentration divided by globulin concentration.Data to assess subsequent reproductive performance for that lactation and milk yield to d 305 of that lactation were then collected during that lactation.
If a cow was blood sampled more than once, only results from the first sampling after each calving were used.A single sample with an implausibly high urea concentration of 86 mmol/L was considered an outlier and removed before statistical analyses.All other biomarker values were considered plausible.
Milk, fat and protein test day yields and lactational yields to 305 d were obtained from DataGene (Bundoora, Victoria, Australia) who manage Australia's national data repository for milk recording data generated by herd commercial milk recording services.Milk, fat Luke et al.: Associations between health biomarkers and fertility and protein yields for the first 120 d of lactation were also calculated for each lactation, as these variables were required to adjust estimated hazard ratios for associations between biomarkers and reproductive performance.These yields to 120 d were calculated from test day data using area under the curve calculations after linear interpolation of daily production between test day values, and assuming that daily production was constant from calving to first test day.Of the 1,121 lactations used to assess the association between each of BHB, NEFA and urea and reproductive performance, 120 d milk yields were available for 1,114 lactations, and these had been calculated using 2 (3 lactations; 0.3%), 3 (46; 4.1%), 4 (543; 48.7%), 5 (505; 45.3%) and 6 (17; 1.5%) test day records.This included the first test day record on or after d 120 of lactation.Calving dates, and artificial insemination and rectal (ultrasound and manual) pregnancy test records were obtained either directly from producers or from DataGene.
Reproductive performance was described for each lactation using the cow's mating start date to conception interval.Right-censored intervals were used for lactations where there was no evidence that the cow had conceived.'Conception' referred only to conceptions detected by pregnancy diagnoses.These intervals were calculated using a 7-step process that commenced with identifying the herd's calving system for the calendar year when the lactation commenced.Three calving systems are used in the Australian dairy industry.In any particular year, a herd can use seasonal calving; all cows calve in a single time period each year, split calving; cows calve in 2 or 3 distinct periods each year, or year-round calving; cows calve in at least 10 mo of the year (Dairy Australia, 2017).To maintain the desired calving system, seasonal and split calving herds have, respectively, one and 2 or 3 calendar dates each year when inseminations commence.These dates are mating start date(s), each being 282 d before the planned start of calving date for the corresponding calving period in the following year.No cows are inseminated before mating start date; after mating start date until the end of that mating period, eligible cows are inseminated.As a result, the period between calving and mating start date varies depending on an animal's calving date.This differs from year-round calving herds where typically mating start date is determined separately for each cow as calving date plus voluntary waiting period (VWP).As the primary reproductive performance objective in seasonal and split calving herds is to achieve conceptions on and shortly after mating start date, the most appropriate reproductive performance metric for individual lactations in seasonal and split calving herds is the mating start date to conception interval.Calving and reproductive data were used to 1) classify each herd's calving system for each calendar year and identify start dates for each calving period for seasonal and split calving herd-years, 2) identify mating start date and end of mating period dates for seasonal and split calving herd-years, 3) for each cow that conceived, estimate her conception date based on rectal pregnancy diagnoses combined with artificial insemination dates, 4) calculate mating start date to conception interval for these cows, 5) for non-conceiving cows, identify their right-censor date, 6) calculate right-censored mating start date to conception intervals for these cows, and 7) identify herd mating periods with incomplete artificial insemination or pregnancy testing data, or both, and implausible apparent conception patterns.For year-round calving herd-years, mating start date was set at 60 d after each cow's calving date.Rightcensor dates were calculated for lactations where the cow had no positive pregnancy diagnoses, for seasonal and split calving herds as the earliest of cull date and end of mating period date or, for year-round calving herds, the earliest of cull date and 364 d after the cow's mating start date.Mating start date to conception intervals were calculated as conception or right-censor date minus mating start date + 1. Mating start date to conception intervals longer than 120 d were right censored at 120 d.
Calving dates, reproductive data and milk production data were available for a subset of study lactations.Numbers and reasons for exclusions and losses to follow-up are detailed in Figures A1, A2 and A3.

Statistical Analyses
All statistical analyses were performed using Stata (versions 16 to 18, StataCorp, USA).
Descriptive analyses.Distributions of biomarkers were described by year within herd i.e., by 'herd-year'.Pearson correlation coefficients for correlations between biomarkers were calculated to inform choices of biomarkers for simultaneous inclusion in multivariable models.These were calculated using the -pwcorr-command.Associated P-values were calculated using the t-distribution.
Associations between biomarkers and reproductive performance.Crude reproductive performance was described using the Kaplan-Meier failure function generated using Stata's -sts list-command.The failure function can be considered as approximating the cumulative percentages of cows that had conceived by time from mating start date.Associations between biomarker concentrations and hazard of conception were assessed using survival analyses of mating start date to conception intervals with Cox proportional hazards models (using Stata's -stcox-command).Hazard of Luke et al.: Associations between health biomarkers and fertility conception is the instantaneous probability of conception for a lactation at any particular time given that the cow had not conceived before that time.Each biomarker was fitted separately as continuous data with the following fixed effects: calving to (blood) sampling interval (continuous data, measured in days), calving to mating start date interval (continuous data, measured in days; linear and quadratic terms fitted), and 2 categorical variables: cow age at calving (2 years (21 to < 34 mo), 3 years (34 to < 46 mo), 4 years (46 to < 58 mo), 5 years (58 to < 70 mo), 6 years (70 to <82 mo), 7 years (82 to < 94 mo) and 8 or more years (94 to < 240 mo)) and the calving system for the herd-year (seasonal, split, or year-round).When analyzing haptoglobin, values recorded as 0 were set at 0.01 and calving system was not fitted, to allow model convergence.
The shape of any relationship between each biomarker and log e (relative hazard of conception) was assessed by fitting fractional polynomials in these Cox models using Stata's -fp-command.For each biomarker, 8 terms (biomarker value raised to powers of −2, −1, −0.5, 0, 0.5, 1, 2 and 3) and 36 combinations of 2 of these biomarker terms were fitted.Following this, the best fitting single term and 2-term models, respectively, the single term and 2-term models with lowest deviance, were identified.The component-plus-residuals (martingale residuals) from the best fitting 2-term models were plotted against the fractional polynomial variable and the plots assessed visually for non-linearity.The overall P-values for testing the null hypotheses that the best fitting single-term and 2-term fractional polynomials are not a better fit than the linear term were also considered, and calculated by comparing deviances between models using likelihood ratio tests.Null hypotheses that there is no linear relationship were assessed using Wald Pvalues for the linear term.Interactions between each biomarker and calving to sampling interval in their joint effects on mating start date to conception interval were also assessed using Wald P-values.Herd was fitted as a shared frailty in all models other than when fitting fractional polynomial terms for urea, globulin and haptoglobin.For these, models with some fractional polynomials would not converge when herd was fitted as a shared frailty.
Adjusted hazard ratios were used to estimate the multiplicative change in daily hazard of conception for each one unit increase in each biomarker concentration.In other words, we estimated the ratio of daily hazard of conception for lactations with any particular biomarker concentration relative to that for lactations where the cow's biomarker concentration was 1 unit less.Failure curves were generated from the Cox models using Stata's -stcurve-command where each cow's biomarker value was set at, variously, the 5th percentile, the median and the 95th percentile for the biomarker (Table 1).When generating these failure curves, all other fixed effects in the Cox model were set at their average value and the frailty was set to one.Impacts of increasing the concentrations of each biomarker on reproductive performance in cows with low values were also estimated as changes in the percentages of cows that had conceived by d 42 and 84 of the mating period from a population where all cows were at the 5th percentile value for the biomarker to populations where all cows were at the median value and 95th percentile value.These were obtained from the failure curves.
Hazard ratios for NEFA, urea, albumin, globulin, calcium, and magnesium were also estimated adjusted for each other; that is, all of these 6 biomarkers were fitted simultaneously as covariates in a Cox model along with the same other covariates as described above.This was done to investigate whether effects of the biomarkers are independent of each other, and thereby assess the validity of using a multi-biomarker panel to better describe and quantify the effects of overall health and nutritional status on reproductive performance without "double counting," by ensuring that the effects of each biomarker are estimated not including effects mediated via others of these biomarkers.These 6 biomarkers were selected from the 9 study biomarkers after consideration of mathematical relationships between biomarkers due to calculation methods, the correlation coefficient estimates for correlations between the biomarkers, and the results from analyses of associations between each biomarker individually and reproductive performance.Albumin to globulin ratio was mathematically related to, and moderately closely and closely correlated with, each of albumin and globulin (correlation coefficient estimates 0.63 and -0.87, respectively) so this ratio was not included and instead albumin and globulin were included as separate variables.β-hydroxy butyrate and haptoglobin were not included as, for each, there was no evidence of an association with reproductive performance when analyzed individually.
A secondary objective was to investigate associations between biomarker concentrations and milk yield.Our reason for doing this was to examine whether any positive associations between biomarker concentration and reproductive performance were explained by reduced milk production.Accordingly, hazard ratio estimates were also estimated after also adjusting, separately, for 120 d milk yield and 120 d milk solids yield i.e., 120 d fat yield plus 120 d protein yield, along with calving to sampling interval, the calving system for the herdyear, calving to mating start date (linear and quadratic terms), and cow age at calving.

Associations between biomarkers and milk yield. Associations between biomarker concentration
Luke et al.: Associations between health biomarkers and fertility and 305 d milk yield were assessed using mixed linear models, using Stata's -mixed-command, with herd and cow within herd fitted as random effects.Calving to sampling interval, interval from herd calving period start date to the cow's calving date (both continuous data, measured in days), and the same 2 categorical variables as for the reproductive performance models, cow age at calving and the calving system for the herdyear, were fitted as fixed effects.For year-round calving herds, the mean interval from herd calving period start date to the cow's calving date for lactations in seasonal and split calving herds was used.The shape of any relationship between each biomarker and 305 d milk yield was assessed by fitting fractional polynomials as described above for log e (relative hazard of conception) in these mixed linear models.Only lactations of at least 120 d duration were included.Maximum likelihood estimation was used.
Interpretation of P-values.We interpreted Pvalues according to the general principles detailed by Wasserstein et al. (2019) and as illustrated by Goodman (1999) in not applying a single P-value threshold but rather placing importance only on low P-values and noting where importance could be attributed based on our results only if further independent supportive evidence are available.

Metabolic Profile Data
Descriptive statistics including the number of lactations analyzed for each biomarker, the number of lactations that had reproductive data, and distributions of each biomarker, are provided in Table 1.The distributions of biomarker concentrations by herd-year are described in Figure A4.For all biomarkers, there was considerable variation within herd-years.For BHB, urea, and albumin, there was also modest to large variation in herd means between years in the same herd relative to the within year variation for the herd.Distributions of calving to sampling intervals, calving to mating start date intervals, and numbers of lactations by cow age and herd-year calving system for lactations included in these biomarker distributions are shown in Tables A1  to A3.
Pearson correlation coefficients for correlations between biomarkers were calculated and are shown in Table 2. Closest correlations were between albumin to globulin ratio and globulin (−0.87), albumin to globulin ratio and albumin (0.63), albumin and calcium (0.46), and albumin and magnesium (0.31).Correlations between all other pairs of biomarkers were weak (<0.30).

Associations between biomarker concentration and reproductive performance
For the 1,121 lactations used to assess the associations between each of BHB, NEFA and urea and reproductive performance, the Kaplan-Meier cumulative failure function, as an estimate of the cumulative percentages of cows that had conceived by time from mating start date, was 51.5% (95% CI not adjusted for clustering by herd or herd-year: 48.6% to 54.5%) at d 42 and 71.3% (95% CI 68.4% to 74.0%) at d 84 of the mating period.Distributions of covariates fitted in the reproductive performance models are shown in Tables A1 to A3. Numbers of lactations and cows included in reproduc- There was no substantial evidence of nonlinear relationships between any biomarker and log e (relative hazard of conception).For each biomarker, from the fractional polynomial plots for the best-fitting 2-term models, (Figure A5), over the range of biomarker concentrations that included most of the observed biomarker values, the component was approximately linearly associated with biomarker concentration.In addition, P-values for comparing each of the best fitting single and 2-term fractional polynomials to the linear term were high (0.126 to 0.984), providing no evidence that there are nonlinear relationships.
Results of Cox models used to investigate associations between biomarker concentrations in the first 30 d after calving and mating start date to conception intervals are reported as (1) adjusted hazard ratios (Table 3) and ( 2) predicted cumulative percentages of cows conceived if biomarker concentrations for all cows in the herd were at the 5th percentile, the median, or the 95th percentile (Figure 1) based on the biomarker values presented in Table 1.P-values for associations between biomarker concentration and hazard of conception were low for albumin, globulin, albumin to globulin ratio, and calcium (Table 3).These associations were positive for albumin, albumin to globulin ratio and calcium in that an increase in biomarker concentration was associated with a higher hazard of conception (and hence, reduced times from mating start date to conception), and negative for globulin, where higher biomarker concentrations were associated with a lower hazard of conception.P-values for NEFA, urea and magnesium were, respectively, 0.057, 0.079 and 0.046, providing reasonable evidence for associations only if further independent evidence are available that supports a negative association for NEFA or, for urea and magnesium, a positive association.The impacts of increasing the concentrations of each biomarker on the reproductive performance of cows with low values were estimated using differences in estimated percentages of cows conceived by d 42 and 84 of the mating period between when the cow's biomarker value was at the 5th percentile and when it was at the 95th percentile (Table A4 and Figure 1).Largest impacts were, in descending order, for albumin (28-29 percentage point increases at both d 42 and 84), albumin to globulin ratio (21 percentage point increases), calcium, globulin, magnesium and urea (both having similar estimated impacts), and NEFA.The marked increase in percentage of cows conceiving on d 36 was because in one herd-year, an ovulation synchronization program was used during the mating period.
Adjusted hazard ratios estimating the strengths of associations between biomarker concentrations and mating start date to conception interval when adjusted for the other biomarkers are shown in Table 4. Estimates for NEFA, urea, albumin and globulin were very similar after this adjustment.But estimated associations were weaker, that is hazard ratio estimates were closer to 1, for calcium and magnesium.
Effects of each biomarker were also estimated after fitting the term for interaction between biomarker concentration and calving to sampling interval (Table A5).There was evidence of interactions between the concentrations of 2 of the biomarkers (urea and globulin) and calving to sampling date, implying that the relationships between these biomarkers and daily hazard of conception differ depending on when samples Only the first samplings in the lactation ≤ 30 d after calving were used (n = 1,094 to 2,106 lactations from 11 to 18 herds other than for the correlations between haptoglobin and each of calcium and magnesium where n = 552 from 7 herds and 448 lactations from 7 herds, respectively).
were taken.P-values for these interactions involving urea and globulin were, respectively, 0.006 and 0.011.For urea, the strength of the association with hazard of conception was weaker (i.e., less positive; estimated hazard ratios closer to 1), with increased time after calving that the sample was taken.In contrast, for globulin, the relationship with hazard of conception was more strongly negative with increased time from calving to sampling.Within biomarkers, hazard ratio estimates were very similar after also adjusting for each of 120 d milk yield and 120 d milk solids (i.e., fat plus protein yields) for the lactation (Table A6).

Associations between biomarker concentration and milk yield
For the 1,806 lactations used to assess the associations between each of BHB, NEFA and urea and 305 d milk yield, mean 305 d milk yield was 7,323 kg (SD 1,831; 5th percentile 2,012; median 6,132; 95th percentile 10,228).Distributions of covariates fitted in the milk yield models are shown in Tables A1 to A3. Numbers of lactations and cows included in milk yield models, respectively, were: BHB, NEFA and urea 1,806 and 1,690; albumin, globulin and albumin to globulin ratio 1,620 and 1,542; haptoglobin 1,040 and 1,033; calcium 1,177 and 1,148; and magnesium 1,105 and 1,093.
Fractional polynomial plots of predicted 305 d milk yields by biomarker concentration are shown in Figure A6.For NEFA, there was evidence of a curvilinear relationship with 305 d milk yields (P < 0.001), for the model with the best fitting 2-term fractional polynomials relative to that with the linear term, the 2 terms being NEFA −2 and NEFA -0.5 (Figure A6).The curvilinear relationship between NEFA and 305 d milk yield was still present after excluding the lactation with lowest NEFA of 0.02 mmol/L (P-value for best fitting 2-term fractional polynomials compared with linear model was still < 0.001; the next highest NEFA concentration was 0.05, observed for 6 lactations).These results showed that cows with very low NEFA concentrations had much lower 305 d milk yields relative to cows with only slightly higher NEFA concentrations.For albumin to globulin ratio and magnesium, P-values for comparing the best fitting 2-term fractional polynomials to the linear term were also low (0.009 and 0.005, respectively).However, from the plots, over the range of most common biomarker values, relationships appeared approximately linear.For all other biomarkers, the corresponding P-values were 0.196 to 0.743, and there was no evidence from the plots of curvilinear relationships over the ranges of most common biomarker values.
For BHB and urea, there was also no evidence of linear relationships with milk yield and P-values for assessing the null hypothesis that there is no linear association were 0.348 and 0.105, respectively.Albumin, albumin to globulin ratio, calcium and magnesium were all linearly positively associated with 305 d milk yields with P-values for linear association relative to null hypothesis < 0.001, < 0.001, < 0.001, and 0.009, respectively.Globulin and haptoglobin were linearly negatively associated with 305 d milk yields and Pvalues for linear association relative to null hypothesis were 0.039 and 0.007, respectively.

DISCUSSION
In this study we investigated the associations between the concentrations of biomarkers in serum measured in the first 30 d after calving, and the reproductive performance and milk production of lactating dairy cows.This period was chosen a) because it is the period in which the vast majority of disease events occur (LeBlanc et al., 2006), and b) because by obtaining information on an animal's health/metabolic state early in lactation Luke et al.: Associations between health biomarkers and fertility The adjusted hazard ratios estimate the multiplicative change in daily hazard of conception for each 1 unit increase in biomarker concentration.Hazard ratio estimates were adjusted for calving to sampling interval, calving system for the herd-year (seasonal, split or year-round), calving to mating start date (linear and quadratic terms), and cow age at calving.Herd was fitted as a shared frailty.

Luke et al.: Associations between health biomarkers and fertility
Figure 1.Predicted failure curves (cumulative percentages of cows conceived) by day of the mating period from Cox proportional hazards models for Holstein cows sampled ≤ 30 d after calving based on serum concentrations of β-hydroxy butyrate (BHB), nonesterified fatty acids (NEFA), urea, albumin, globulin, albumin to globulin ratio, haptoglobin, calcium and magnesium.Blue, gray and orange lines represent predictions if the cow's value for the biomarker was, respectively, the 5th percentile, median and 95th percentile of the values for all lactations (Table 1).For these predictions, all other fixed effects in the Cox models were set at their average value and the frailty was set to one.Numbers of lactations and herds used for analyses for each biomarker are shown in Table 3.
provides producers with sufficient time to implement nutritional or management interventions to improve subsequent performance.Discussion of possible interventions is beyond the scope of this study but is the focus of ongoing work by our research group.To date, most studies examining associations between biomarker concentrations and reproductive performance have been undertaken in year-round calving herds, often in cows being fed in a TMR system.The current study was conducted in 24 pasture-based herds.To the best of the authors' knowledge, this is the first such study to be undertaken in Australia, and one of only very few studies to examine the associations between reproductive performance and milk production and a suite of biomarkers rather than just one or 2 biomarkers.

Associations between biomarker concentrations and reproductive performance
Immune status.Perhaps the most interesting finding of this study was the strong positive association between serum albumin concentration and reproductive performance, as measured by mating start date to conception intervals.Our results are consistent with those from work undertaken by researchers from Institute for Research on Animal Diseases at Compton in the UK who pioneered the use of serum metabolic profiling in dairy cattle.For example, this group demonstrated that serum albumin concentration between 40 and 100 DIM was inversely related to the number of services per conception (Rowlands et al., 1977), and that cows requiring 4 or more services had lower serum albumin concentrations between weeks zero and 2, and 7 and 9 of lactation, compared with cows that conceived at first service (Rowlands and Manston, 1983).Interestingly, we could find no recent references to associations between albumin and reproductive performance.Albumin, the most abundant plasma protein in mammals, is synthesized primarily in the liver and is involved in many important biological functions including maintaining circulating blood volume and oncotic pressure, and acting as a carrier for hormones, fatty acids, metabolites and ions (Majorek et al., 2012).Albumin is also a negative acute phase protein (concentrations decrease in response to inflammation; Jain et al., 2011) and is used as a biomarker of both chronic inflammatory disease, and long-term protein intake (Don et al., 2004).Given that albumin is involved in such a wide range of biological functions, further work is required to a) understand the underlying mechanism(s) driving improved reproductive performance in cows with higher albumin concentrations in early lactation, and b) what factors are influencing albumin concentration in early lactation.This is especially true in pasturebased systems where less is known about determinants of albumin concentration than in TMR systems.Importantly, our results showed that the strength of the association between albumin concentration and reproductive performance increased with sampling time after calving, and we believe that this finding warrants further investigation.
The ratio of albumin to globulin, which is used as a non-specific indicator of innate immune status (Piccinini et al., 2004), was also strongly positively associated with reproductive performance.These results are consistent with a recent study that demonstrated that cows classified as having a high albumin to globulin ratio before calving had fewer days open and had fewer services per pregnancy than cows with either low or intermediate ratios (Cattaneo et al., 2021).Our results therefore support the hypothesis offered by those authors that albumin to globulin ratio is a promising prognostic biomarker for transition cow health and fertility.Again, further work is required to understand the biological mechanism(s) driving these associations.Previous work undertaken by our group indicates that there may be a genetic component to this association as albumin to globulin ratio is a heritable trait in Holstein cattle (0.20 < h 2 < 0.41; Cecchinato et al., 2018, Luke et al., 2019a), and the correlation between genomic estimated breeding values for daughter fertility and albumin to globulin ratio, which is indicative of the genetic correlation, is 0.30 (Luke et al. 2019a).There-  Adjusted only for calving to sampling interval, calving system for the herd-year (seasonal, split or year-round), calving to mating start date (linear and quadratic terms), and cow age at calving.Herd was fitted as a shared frailty.
3 Adjusted for the same covariates listed above and all other biomarkers listed in this table.Herd was fitted as a shared frailty.
fore, genetic selection for cows with higher albumin to globulin ratios could lead to not only improved animal health and immune function, but also better fertility.Energy balance.We found no evidence that serum BHB concentration was associated with reproductive performance in the study population.β-hydroxy butyrate is a commonly used biomarker of negative energy balance and ketosis (Ospina et al., 2010a), and several studies from a diverse range of production systems have shown that elevated serum concentrations of BHB in early lactation are negatively associated with subsequent fertility.For example, Compton et al. (2014) showed that grazing dairy cows in New Zealand that had BHB concentrations >1.2 mmol/L any time in the first 5 weeks of lactation had a lower pregnancy rate at wk 6 of the mating period.Ospina et al. (2010a) showed that TMR-fed cows with BHB concentrations ≥10 mg/ dL between 3-and 14-d postpartum had a 13% lower hazard of pregnancy.One possible explanation for our differing results is that very few animals in our data set had high BHB concentrations; BHB concentration at the first sampling in the lactation was greater than 1.2 mmol/L for only 27 cows (1.3%).This is considerably lower than hyperketonemia prevalences using similar cut-points of 43.2% 3 and 16 DIM reported in TMR-fed cows in the USA (McArt et al., 2012), 16.8% in grazing cows between 7 and 12 DIM in New Zealand (Compton et al., 2014), and 21.8% between 2 and 15 DIM in Europe (Suthar et al., 2013).The fact that we included data from cows up to 30 DIM may also help to explain our generally low BHB concentrations, as prevalence of hyperketonemia has been shown to peak at 5 DIM and decrease rapidly thereafter (McArt et al., 2012).In the current study, relatively few cows were sampled by d 5 post calving but modest proportions were sampled between d 5 and 11; 5th and 25th percentiles of calving to sampling interval distributions were 4-5 d and 10-11 d, respectively.Thus, the absence of evidence that serum BHB concentration is associated with reproductive performance from the current study may have been due to the limited range of BHB values in study cows.This absence of evidence would also be as expected if higher BHB concentrations are less indicative of poor reproductive performance when BHB is measured longer after calving.Our point estimate for the interaction term was imprecise but, assuming interaction of that magnitude, point estimates of hazard ratios for the effects of BHB when the cow was sampled 5, 15 and 25 d after calving were 1.33, 1.05, and 0.84, respectively, so were consistent with this expectation.Further work is required to better understand the epidemiology of hyperketonemia and its association with reproductive performance in Australian dairy herds.
Excessive negative energy balance in the postpartum period is known to have a negative impact on reproductive performance in dairy cows (Wathes et al., 2007).Given that NEFA is used as an indicator of fat mobilization, and therefore the magnitude of negative energy balance, it makes sense that in our study, cows with higher NEFA concentrations had poorer reproductive performance.This finding is also consistent with Ospina et al. (2010a), who demonstrated that cows with relatively high serum NEFA concentrations had a lower risk of pregnancy.
Protein nutritional status.Our results indicated that cows with higher serum urea concentrations in the first 30 d after calving were more likely to conceive than cows with lower urea concentrations.This is in stark contrast to many published studies where both milk and blood urea concentrations were negatively associated with reproductive performance in cows fed a TMR.For example, Butler et al. (1996) found that plasma urea concentrations >6.8 mM on the day of AI were associated with an approximate 20% decrease in pregnancy rate to that AI.From a metanalysis, Raboisson et al. ( 2017) concluded that serum urea concentrations above 7.0 mM are associated with a 43% lower odds of conception compared with cows with lower urea concentrations.One possible explanation for this discordance in results could be that cows included in the current study were all managed in pasture-based systems, and had therefore adapted to high rumen degradable protein levels in the diet (Laven et al., 2007).Furthermore, milk urea nitrogen, which is linearly correlated to serum urea concentration, has been shown to be an indicator of dietary crude protein intake (Roseler et al., 1993, Nousiainen et al., 2004, Macrae et al., 2006).Therefore, in grazing systems, urea may be an indicator of feed intake, which has been shown to be positively associated with reproductive performance (Lucy et al., 1992).
Macromineral status.We note with interest that the impact of the association between magnesium concentration and mating start date to conception interval is almost identical to that for the association between urea concentration and mating start date to conception interval.Serum magnesium concentration varies with dietary intake of magnesium, and is sometimes used as an indicator of dry matter intake when the concentration of magnesium in a diet is known i.e., in a TMR system (Macrae et al., 2006).At first inspection, this appears to support the hypothesis that urea may also be an indicator of feed intake.However, the correlation between urea concentration and magnesium concentration was weak (Pearson's r = 0.07), suggesting that the associations between each of these 2 biomarkers and reproductive performance are mediated through differ- Luke et al.: Associations between health biomarkers and fertility ent biological mechanisms.Further work is therefore warranted to elucidate the mechanisms driving these associations.
We observed a strong favorable association between calcium concentration and mating start date to conception interval.Our results are consistent with the findings of several studies which have reported positive associations between serum calcium concentration and reproductive performance.For example, cows with higher serum calcium levels had shorter intervals from calving to first estrus (Rodriguez et al., 2017), and from calving to first service and conception (Mahen et al., 2018).Also, Chapinal et al. (2012) showed that the odds of pregnancy at first AI were lower in herds where more than 25% of cows had serum calcium concentration less than 2.1 mmol/L in the first week of lactation.Interestingly, Neves et al. (2018a, b) showed that the sequelae of hypocalcemia depended on the time at which a hypocalcemic event occurred.In our data set, there was no interaction between calcium concentration and calving to sampling date.As with BHB, our point estimate for the interaction term between calcium concentration and calving to sampling date was imprecise but, assuming interaction of that magnitude, point estimates for hazard ratios for the effects of calcium when the cow was sampled 5, 15 and 25 d after calving were 1.54, 2.03, and 2.67, respectively, so were consistent with much larger effects of calcium when measured longer after calving.However, very few of our samples were taken within 48 h of calving, which is known to be the period of highest risk for hypocalcemia in dairy cows (Goff 2008).Therefore, as with BHB, we believe more samples collected in the immediate postpartum period are required to better understand the association between serum calcium concentration and reproductive performance.
The marked increase in percentage of cows conceiving on d 36 was because in one herd-year, an ovulation synchronization program was used during the mating period.That increase would not have affected our results as we used a semi-parametric model so the baseline hazard was determined by the data and not by imposing a parametric distribution on the data (that would inevitably not fit these data due to that increase).In addition, that management strategy was imposed on all cows in that herd-year regardless of their biomarker status, thus that management strategy would not have influenced the fitted relative hazards of conception.

Impact of milk yield on the observed associations between biomarker concentration and reproductive performance
Our hazard ratio estimates for associations between biomarker concentration and reproductive performance were very similar after also adjusting for each of 120 d milk yield and 120 d milk solids, indicating that our observed relationships were not due to any differences in milk yield between cows with low and high biomarker concentrations.Further, of the biomarkers that were positively associated with reproductive performance or where there was some evidence for this (urea, albumin, albumin to globulin ratio, calcium, and magnesium), all except urea were also positively associated with milk yield, and no relationship was evident for urea.For example, cows with higher albumin concentrations had both shorter mating start date to conception intervals and higher 305 d milk yields relative to cows with lower albumin concentrations.Relationships were also concordant for globulin; cows with higher globulin concentrations had longer intervals from mating start date to conception intervals and lower 305 d milk yields.These results suggests that, in the Australian context, strategies to increase reproductive performance associated with improvements in these biomarker concentrations may be successful without causing decreases in milk production.
Nonesterified fatty acids was the only biomarker for which these relationships were discordant, with cows with higher concentrations having higher milk yield but no increase in (and possibly poorer) reproductive performance.Thus, cows in this study with very low NEFA concentrations tended to have lower milk yield and similar or possibly better reproductive performance than cows with higher NEFA.Given this combination, it is possible that the cows with lower NEFA concentrations may have been animals that were not mobilizing body fat reserves rather than poor condition cows with no reserves to mobilize.If the latter scenario was the situation, that is cows with low NEFA concentrations being under-conditioned, we would have expected very low NEFA concentrations to be associated with lower, rather than similar or possibly better, reproductive performance as observed.

Practical applications
We observed marked differences between biomarkers in the extent of variation between and within herdyears.For example, urea concentration varied considerably between herd-years relative to within herd-year variation.This suggests that herd-level nutritional or management interventions, or both, may be available Luke et al.: Associations between health biomarkers and fertility that increase average urea concentrations of some herds.Further, as most biomarkers had large withinherd variation, investigations are warranted to assess whether subgroups of animals can be identified, and reproductive performance improved through targeted interventions.
Many studies investigating associations between biomarker concentrations and reproductive performance, milk production, or health events, have proposed concentration thresholds above or below the point at which adverse events are likely to occur.Defining thresholds can be useful for interpreting individual animal diagnostic tests to identify sick versus healthy animals.This binary classification approach can also be used to assess herd status, and it is possible to calculate the number of individual animals that need to be tested to provide a sufficiently precise estimate of the overall health of the herd.In other words, this binary approach can be used to estimate the proportions of tested animals where biomarker concentrations are beyond threshold levels, and thus identify whether there are causes for concern at herd level (Ospina 2010).Collectively, our results do not support the hypotheses that there are thresholds in the relationships between biomarkers and log e (relative hazard of conception).That is, there do not appear to be threshold values for each biomarker below which there is no important relationship with mating start date to conception interval and above which there is also no important relationship with mating start date to conception interval.Therefore, by defining thresholds and collapsing continuous biomarker concentrations into binary traits, we risk losing important information about associations, as the relationships are present both below and above the thresholds.This implies that using the threshold approach may be less useful for modeling and or monitoring the impact of management or nutritional interventions at herd level.A method that captures the continuous nature of these relationships to inform management and or nutritional interventions is therefore required.Having herd-level data is particularly important in seasonal and split-calving systems where 1) reproduction has a very large impact on farm profitability (Malcolm 2017), and 2) large numbers of animals are calving in a short period of time so making decisions at an individual animal level is much more difficult than in herds using year-round calving.
Another potential application of this work is to use hazard ratios to construct dose response curves to help quantify the economic impact of nutritional and management changes on the reproductive performance of a herd or group of animals.Large studies are required to meaningfully assess impacts of nutritional and management interventions on reproductive performance using experimental approaches.Intensive, well-designed ex-periments are extremely costly to run, and, to implement such experiments on a sufficient scale to provide meaningful results on any impact(s) on reproductive performance is often cost prohibitive.However, if serum concentrations of biomarkers are measured in animals under different treatments, and if relationships between biomarker concentrations and reproductive performance have been estimated from observational studies such as the current studies, these data can collectively be used to estimate the effects of a treatment on reproductive performance.
The estimated strengths of relationships between each of albumin and globulin and mating start date to conception interval did not change substantially when the effects of each of these biomarkers were estimated adjusted for the concentration of 5 other biomarkers.This indicates that the effects of each of these 2 biomarkers are independent of effects for the other 5 biomarkers, and that a panel of multiple biomarkers may collectively predict reproductive performance better than any single biomarker.It was beyond the scope of this study to assess the abilities of multiple biomarkers in predicting subsequent reproductive performance, but this work is currently underway.
Mid-infrared (.MIR) spectroscopy of milk, undertaken as part of routine milk recording, has provided promising predictions of serum urea concentration in dairy cows (Luke et al., 2019b, Ho et al., 2021).Given the ease and cost-effectiveness of MIR predictions in milk, results of the current study suggest a potential role for MIR-predicted urea as a management tool for producers to assess dry matter intake, which is known to be an important driver of both milk production (Roche 2007) and reproductive performance (Lucy et al., 1992).Given the strong phenotypic and genetic correlations between serum urea concentration and MIR-predicted serum urea concentration (van den Berg et al., 2021), the latter may also be a useful indicator trait for genetic selection for feed efficiency traits such as residual feed intake.

CONCLUSIONS
We investigated the relationships between a suite of biomarkers and each of reproductive performance and milk production in pasture-based dairy cows in southeastern Australia.We found associations between several biomarkers and reproductive performance, defined as mating start date to conception interval.Of the biomarkers investigated, serum albumin concentration and serum albumin to globulin ratio had the largest impacts on reproductive performance in cows with low concentrations of the biomarker.Associations with milk yield were concordant with associations with reproductive Luke et al.: Associations between health biomarkers and fertility performance for all biomarkers except NEFA.For all biomarkers, we observed substantial variation within and for some biomarkers, between herd-year groups, indicating that there may be scope to improve biomarker concentrations and reproductive performance and milk yield through strategies such as nutritional and management interventions, and through genetic selection. 2 . Approval to proceed was granted by the Agricultural Research and Extension Animal Ethics Committee of the Department of Energy, Environment and Climate Action (DEECA, 475 Mickleham Road, Attwood, Victoria 3049, Australia), and the Tasmanian Department of Primary Industries, Parks, Water and Environment (DPIPWE Animal Biosecurity and Welfare Branch, 13 St Johns Avenue, New Town, Tasmania 7008, Australia).Project approval numbers 2017-05,

Table 1 .
Luke et al.:Associations between health biomarkers and fertility Descriptive statistics of calving to sampling intervals and serum biomarker concentrations within 30 d after calving in pasture-based Holstein cows 1 performance models, respectively, were: BHB, NEFA and urea 1,121 and 1,103; albumin, globulin and albumin to globulin ratio 946 and 944; haptoglobin 634 and 632; calcium 837 and 837; and magnesium 765 and 765. tive

Table 2 .
Luke et al.:Associations between health biomarkers and fertility Pearson correlation coefficients for correlations between serum biomarkers in pasture-based Holstein cows 1 .P-values for testing the hypotheses that r = 0 are shown in brackets; P < 0.002 if no P-value shown

Table 3 .
Associations between serum biomarkers measured ≤ 30 d after calving and mating start date to conception intervals in pasture-based Holstein cows

Table 4 .
Luke et al.:Associations between health biomarkers and fertility Associations between biomarkers measured ≤ 30 d after calving and mating start date to conception intervals in pasture-based Holstein cows adjusted for other biomarkers 1 .The adjusted hazard ratios estimate the multiplicative change in daily hazard of conception for each 1 unit increase in biomarker concentrationThe same 765 lactations from 765 cows and 11 mating periods from 10 herds were used to generate all results reported in this table.