Bovine mastitis is a multietiologic disease that can be induced by more than 130 different species of microorganisms. Mastitis pathogens have traditionally been classified as "minor" or "major" pathogens in order to assess the prognosis of the infection and to establish priorities in the adoption of preventive and control measures. Thus, major pathogens are related to a higher impact on cattle udder health, as well as on milk quality and productivity ( Zadoks & Fitzpatrick, 2009 ).

This classification has generally been based on the frequency of presentation in clinical mastitis as well as on the somatic cell counts (SCC) in subclinical mastitis, since it is broadly considered that the existence of an intramammary infection (IMI) is the most important cause of an SCC increase ( Dohoo & Meek, 1982 ). Different thresholds have been set to SCC to consider the existence of IMI in subclinical mastitis (100,000-500,000 cells/mL), but 200,000 cells/mL is the most widely used ( Schwarz et al. , 2010 ; IDF, 2013 ). Ho­wever, since SCC also vary with the lactation stage, age, stress of the animals, season and time and frequency of milking ( Dohoo & Meek, 1982 ; Harmon, 1994 ), this threshold may inaccurately classify some mastitis ( Schwarz et al. , 2010 ).

Some microorganisms like Streptococcus agalactiae, Streptococcus dysgalactiae, Streptococcus uberis, Sta­phy­lococcus aureus, Escherichia coli or Klebsiella pneumoniae are generally considered in literature as major pathogens ( Bradley, 2002 ; Zadoks & Fitzpatrick, 2009 ), whereas coagulase-negative staphylococci and Corynebacterium spp. are considered minor pathogens ( Bradley, 2002 ; Pyörälä & Taponen, 2009 ). However, there are a number of pathogens which regularly present low rates of isolation ( Pitkälä et al. , 2004 ; Tenhagen et al. , 2006 ) and have not been reported in any of those groups (e.g., Streptococcus spp. or Enterococcus spp., as well as yeasts, coliforms or other Gram-negatives, etc.) ( Bradley, 2002 ). Consequently, when these pa­thogens are studied, they are frequently merged in groups with other bacteria on the basis of genus, family or microbiological group. Thus, their effects on cattle mastitis are considered as a group or assumed to be similar to other more well-known pathogens of the same genus or microbiological group ( Djabri et al. , 2002 ; de Haas et al. , 2004 ; Malinowski et al. , 2006 ; Schwarz et al. , 2010 ). Nevertheless, these groups may not be so homogenous to make such assumptions. For example, non- aureus staphylococci (NAS) are historically studied as a group in cattle mastitis and, while different authors agree that SCC in IMI caused by NAS as a whole are lower than that when caused by Staph. aureus ( Schukken et al. , 2009 ; Supré et al. , 2011 ; Taponen et al. , 2007 ), results are contradictory when different NAS species are individually compared. Therefore, several authors did not find any difference among NAS regarding the severity of clinical signs ( Jarp, 1991 ; Taponen et al. , 2006 ) or SCC ( Thorberg et al. , 2009 ), while others did find statistically significant differences ( Sampimon et al. , 2009 ; Fry et al. , 2014 ; Condas et al. , 2017 ), and some NAS species (e,g. Staphylococcus chromogenes, Staphylococcus si­mu­lans and Staphylococcus xylosus ) even presented similar SCC to Staph. aureus ( Supré et al. , 2011 ).

Pathogens with low rates of isolation in mastitis gain importance by summing all the cases in which they all are involved ( Makovec & Ruegg, 2003 ; Piepers et al. , 2007 ; Schwarz et al. , 2010 ; Kalmus et al. , 2011 ). Besides, mastitis control programs have decreased the prevalence of some pathogens, such as Staph. aureus or Strep. agalactiae , but also increased the relative rate of isolation of others, which were less frequent in the past ( Zadoks & Fitzpatrick, 2009 ; Fernández et al. , 2013 ). Consequently, it is necessary to gain solid knowledge about the relative behavior of these pathogens in cattle mastitis.

Thus, the aim of this study was to analyze the behavior of a wide spectrum of mammary pathogens regarding different variables: percentage of isolation in clinical mastitis, SCC in non-clinical mastitis and percentage of isolation in samples with SCC ≤ 100, 000 cells/mL. These results were subsequently used to classify the different mammary pathogens into homogenous groups.

A total of 240,232 samples were submitted to the laboratory during the period of study; 32,852 (13.7%) were contaminated, 30,561 (12.7%) did not show any growth, and more than one species were isolated in 1,854 (0.8%) samples. Consequently, these samples were removed from the study. Additionally, 1,620 were also removed because the quantity of the sample was insufficient. As a result, 173,345 samples were finally included in this study; 50,389 (29.1%) were classified as clinical mastitis and 122,956 (70.9%) as non-clinical. Among the samples categorized as non-clinical mastitis, 8,036 (6.6%) presented SCC ≤ 100,000 cells/mL.

Forty-three different species were identified. Ho­wever, 5.4% of Gram-positive catalase-positive cocci could not be identified to the species level, as well as 1.9% of Streptococcus spp. different from Strep. agalactiae , 2.2% of Enterococcus spp. and 1.3% of Gram-negative. In 20% of yeasts, the identification to, at least the genus level, could not be achieved either.

Pathogens whose isolation frequency was over 0.1% accounted for a total of 163,741 (94.5%) samples. They represented 94.2% of the non-clinical samples with SCC > 100,000 cells/mL (n = 108,226/114,889), 95% of the non-clinical samples with SCC ≤ 100,000 cells/mL (n = 7,714/8,120), and 94.8% of the clinical samples (n = 47,719/50,336). The SCC and LS values of these pathogens are presented in Table 1, which shows data from the SCC > 100,000 threshold. Table 2 shows the %ICS and %ISU100 for each microorganism identified.

Principal component analysis prior to clustering provided one first component (Dim 1) which accoun­ted for 87.46% of the variance. The observed varia­bles contributed to this component similarly (30.84%-36.15%), being positively correlated in the case of mLS and %ICS and negatively in %ISU100. The second component (Dim 2) accounted for a 10.09% of the variability, highlighting the contribution of %ICS (61.05%) and, to a lesser extent, %ISU100 (36.97%), although the correlations were low (Table 3).

Figure 1 shows the dendrogram from the cluster analysis. Looking at the results in the dendrogram, the optimum solution was considered when a cut-off threshold at a height of 3.88 was selected, resulting in the existence of four groups. The distribution of these clusters according to the coordinates of the microorganisms in the two main components is shown in Fig. 2a. Pairwise plots between the observed values of the three variables (mLS, %ICS and %ISU100) can be observed in Figs. 2b, 2c and 2d. In general, there is a good agreement between the distribution of the microorganisms and the assigned clusters, especially, between %ICS and %ISU100 (Fig. 2d).

Table 4 shows the significance of the variables in each group in regard to the overall mean. Group I was characterized by low values for mLS and %ICS and high values of %ISU100. Group III presented high values of mLS and low values of %ISU100. Group IV presented the highest values of mLS and %ICS and lower values for %ISU100. Group II did not show any differences regarding the overall mean. Therefore, considering the results of the other groups, it can be placed between Group I and III.

The maximum α error associated with these groups can also be observed in Fig. 1 and, overall, it matches rather well with the results of the SP analysis.

In general, the groups obtained here were quite consistent with previous literature ( Djabri et al. , 2002 ; Taponen et al. , 2007 ; Schukken et al. , 2009 ; Kalmus et al. , 2011 ; Supré et al. , 2011 ) and they may provide useful insight about pathogens with low isolation ra­tes regarding other well-known microorganisms. This classification aims to reflect the global trends in the population of the pathogens regarding mLS, %ICS and %ISU100, which is useful to assess the diagnostic significance of an isolation. We expected high mLS and %ICS values in pathogens related to severe masti­tis, whilst these values should decrease in pathogens associated to less severe mastitis, presenting high %ISU100 in this case. Pathogens were mainly ordered in that way and, according to PCA, the resulting groups were principally formed based on the values in Dim 1 (Table 3), as it can also be interpreted from the V-test results (Table 4). However, some of them punctually diverged from this order, principally due to the expression of different mLS values from those that could be expected from their %ICS values (Fig. 2b). This shows that some variations with regard to this general tendency are also possible, suggesting a likely more complex epidemiology underlying these results.

Group I contained most of minor pathogens, such as Corynebacterium spp. and most of NAS. Pathogens in this group showed the highest %ISU100 and the lowest mLS. Actually, isolations of some species in this group, like Corynebacterium bovis , have proved to mostly reflect a colonization of the teat channel rather than a real IMI ( Bexiga et al. , 2011 ). However, major pathogens were divided in three groups (II-IV), which seem to indicate certain heterogeneity among them. Group II presented the least virulent one, and it included pathogens like Staph. aureus and Staph. intermedius , whereas Strep. agalactiae, Strep. uberis, Strep. dysgalactiae were classified in Group III. Group IV, the most virulent one, included pathogens like E. coli and Klebsiella pneumoniae .

Staph. intermedius was the only NAS in the same group as Staph. aureus (Group II), suggesting a higher pathogenicity of Staph. intermedius in comparison to the rest of NAS. As aforementioned, contradictory findings are reported in NAS mastitis ( Jarp, 1991 ; Taponen et al. , 2006 ; Sampimon et al. , 2009 ; Thorberg et al. , 2009 ; Supré et al. , 2011 ; Fry et al. , 2014 ; Condas et al. , 2017 ). These may reflect differences in NAS isolation frequencies or in the pathogenicity of the strains among the different studied populations, which include, in some cases, a limited number of infections.

Group III included pathogens known to be associated to severe mastitis, but also microorganisms with a discussed role in cattle mastitis, like yeasts or Prototheca spp. Yeasts are generally assumed to provoke weak infections rarely associated with clinical mastitis and likely to self-cure ( Richard et al. , 1980 ). However, outbreaks with high SCC have also been reported (Malinowsky et al. , 2006) and our results seem to be more consistent with this finding. Likewise, mastitis caused by Prototheca spp. is also considered to be rarely associated with clinical signs ( Scaccabarozzi et al. , 2008 ), but we also found this genus in Group III. This could indicate that these kinds of infections are responsible for mastitis of relative importance. It is also noteworthy that Pseudomonas aeruginosa , described as a cause of chronic mastitis which can also provoke either acute or subacute processes or subclinical disease ( Hogan et al. , 1989 ), was classified in Group III in this study along with other pathogens responsible of severe mastitis.

Group IV was formed by mammary pathogens rarely isolated in mastitis samples such as T. pyogenes and Nocardia spp. This was not an unexpected finding since both pathogens had already been associated with severe mastitis ( Packer, 1977 ; Tarabla et al. , 1993 ). However, according to our results, Pasteurella multocida and Strep. hyointestinalis could also be associated with processes of similar severity.

It is interesting that despite pathogens with %ISU100 >5% were included in Group I, several pathogens included in Groups II and III presented %ISU100 over 2%. Ranges and means of SCC in IMI vary among different pathogens ( Djabri et al. , 2002 ), so their isolation frequency is also expected to vary in samples with low SCC. The diagnostic importance of isolations in samples with low SCC may vary depending on the pathogen. Thus, pathogens of Group I in samples with low SCC may indicate mild disease or colonization without IMI, but major pathogens isolated in this type of samples may be interpreted as an early indication of IMI ( Schwarz et al. , 2010 ) and their detection would be of importance in cattle mastitis programs. Moreover, higher %ISU100 were found in pathogens usually transmitted from other infected animals (e.g., Staph. aureus and Strep. agalactiae ) than in pathogens with an environmental origin (e.g., E. coli or Strep. uberis ) (Table 2). This fact has a great importance for control, thus calling for the implementation of new biological markers in order to get a more accurate information of the actual udder health status of dairy cows ( Damm et al. , 2017 ).

It is also noteworthy that pathogens traditionally merged in microbiological groups were separated in different clusters in our study, showing potential variability in their pathogenic role (Fig. 1). This could explain some of the divergences found in the literature for species of the genera Enterococcus, Aerococcus or Lactococcus . Species within these genera still have a discussed role in cattle mastitis ( Devriese et al. , 1999 ; Zadoks et al. , 2011 ) and while some authors consider them to be minor pathogens ( Guélat-Brechbuehl et al. , 2010 ), outbreaks of clinical mastitis involving some of them have also been reported ( Todhunter et al. , 1995 ). This kind of differences had already been described for some pathogens among the Enterobacteriaceae family. For example, E. coli and Klebsiella spp. had been attributed to the most severe cases, whereas less clinical severity had been observed in infections caused by Serratia spp. and Enterobacter spp., which were more frequently related to chronic mastitis ( Schukken et al. , 2012 ). Since frequencies of isolation of the species that compose these groups may vary among studies and these pathogens may be implicated in mastitis of different severity, our results suggest that merging these species may lead to incorrect inferences of their role in cattle mastitis.

It has to be noted that it is impossible to evaluate all the factors related to the severity of mastitis in this kind of study (such as the duration of the infection or the severity of the exhibited signs in clinical mastitis). Cattle mastitis is dependent on different factors that may affect to the presentation and severity of clinical signs and SCC. Differences in pathogenicity have been observed among strains in both SCC ( Zadoks et al. , 2002 ) and clinical outcomes ( Middleton & Fox, 2002 ). In addition, there are also cow-specific characteristics that may lead to wide variations in the course of IMI such as parity, month of lactation or SCC in previous lactation ( Steeneweld et al. , 2008 ; Oliveira et al. , 2015 ). Including cow and pathogen factors in the cluster analysis would have provide a more robust result and it would be necessary to a more complete and accurate determination of the pathogenicity of these pathogens, but unfortunately, we could not register animal or herd level information in this study and so, a certain degree of bias is expected. However, since we have analyzed a large number of isolations by pathogen we aimed to be representative of the global trends of each pathogen in a population. Nevertheless, it must be noted that a different cluster solution may be expected in other populations due to differences of cow-speci­fic factors or in the circulating strains. Despite that, our classification is pretty consistent with previous literature ( Djabri et al. , 2002 ) and it allows the relati­ve comparison among pathogens. This is particularly interesting for microorganisms with little isolations since their available epidemiological information is limited and it is frequently derived from few herds, which may be influenced by particular characteristics of the animals and herds analyzed. An overrepresentation of severe mastitis in samples submitted for diagnosis is also expected in comparison to a random sampling, so it is possible a trend to over-associate microorga­nisms to more severe cases. Actually, and probably as a consequence of the design of our study, our SCC values were higher than those reported by Djabri et al. (2002) for the pathogens they included in a meta-analysis to evaluate the associated effects of bacteria or group of bacteria with IMI. However, in our opinion, this does not affect our aims, since this study does not intend to establish concrete values for each pathogen, but to perform a relative comparison between the analyzed pathogens.

This study provides a picture of some of the outcomes related to the pathogenicity of several pa­thogens involved in cattle mastitis. Our results are also helpful for the understanding of the importance and frequency of isolation of the different pathogens in samples with low SCC. The groups obtained in this study can aid, together with other factors, in the classification of mammary pathogens. The classi­fication of microorganisms is more complex than the traditional differentiation between minor and major pathogens, especially in the latter case. Our results provide a starting point for the interpretation of the information of pathogens with low prevalence, and highlight the differences among pathogens which are frequently merged in microbiological groups for their study. Further studies, in which other epidemiologic characteristics are added, could aid in understanding the differences in the epidemiology of mastitis pathogens.