Animal & Veterinary
Antimicrobial Resistance in Historical Salmonella Typhimurium Using High-Density Microarrays by Heather Harbottle, Ph.D.
DR. HARBOTTLE: Okay. I am getting this all open here. Just while I am opening, I wanted to thank the organizers for giving us the opportunity to talk about our research projects today. It is very interesting and exciting to hear about the programs in other countries. I think it is also very important to talk about the molecular research that we do in support of these programs. So with no further adieu, I will begin.
The project that I would like to talk about today is actually a project that was funded externally to NARMS. It was funded by the Critical Path Initiative at the FDA and it is a collaborative project between myself from CVM and also a collaborator from CFSAN named Scott Jackson. So I will be telling you a little bit more about our particular roles in this project as I go forward.
What our objective was was to use a historical as well as a contemporary collection of Salmonella enterica serotype Typhimurium in order to look at emerging resistance in antimicrobial resistance phenotypes and genotypes using a custom high-density microarray. These were all from human infections.
So just a little bit about our materials and methods. The first set of isolates that we focused on were a collection from the ATCC that we affectionately call our historical collection. These isolates were obtained from the ATCC including some antimicrobial resistance data. They encompassed a time period from 1948 to 1995.
After that point, we asked our collaborators at CDC, specifically Dr. Jason Folster and Dr. Jean Whichard, if they would contribute some strains to this collection as we were only looking at human infections. So they very kindly sent us some strains from 1996 to 2006 to complete the study.
For the susceptibility typing, broth microdilution was used with the common NARMS plates, so I will not dwell on that much. But for the microarray, I will go into the microarray design a little bit in the next slide, but we used standardized protocols that were developed at CFSAN.
The microarray that I am going to be discussing with you is called the FDA Affymetrix Salmonella enterica/Escherichia coli microarray or commonly called FDA SEEC. The protocols and scanning protocols were standardized in CFSAN and were followed accordingly.
For data analysis, gene present and absent calls, so this was purely gene detection. This is not an expression experiment. Our gene present/absent calls were determined using the modified version of the Affymetrix MAS5 algorithm. Our probe set data was summarized using RMA or robust multiarray averaging.
So this is the design of the microarray and this is a little bit of information about the microarray. It is a bit different from what Dr. Lindsey was talking about a minute ago. It is high-density, photolithographic microarray, so it is much higher density as you can see.
It was designed in 2008 largely by our collaborators at CFSAN based on some work that they had done previously. Their main objective and our main objective obviously overlapped quite a bit with the antimicrobial resistance elements which were largely contributed by CVM from our in-house array, but also some other databases were mined, including NCBI as well as the Lawrence Livermore database for resistance genes as well as virulence genes.
So that is a main portion of what we will be focusing on today. However, the array also includes all known sequence chromosomes at the time. So in 2008, that included a number of E. coli, Shigella, and for our interest today, included the Salmonella enterica strains that were sequenced at the time which were 38. So this boils down to a total of about 83,000 genes or 2.5 million probe sets
So first I would like to sort of give you an overview of what we were looking at for percent resistance of the entire collection. There was 145 strains and as we said from the two different sources, ATCC and our NARMS CDC partners.
Overall, the collection was largely susceptible, about 43.5 percent were susceptible. However, we did see a number of different resistance patterns. The ACSSuT pattern that we have been talking about at some length today was around 15 to 17 percent. Relatively high levels of resistance to the older drugs such as ampicillin, tetracycline, streptomycin, sulfamethoxazole. A few that were resistant to four or more drugs, somewhere around the 23 percent mark, and a very little level of nalidixic acid and zero Cip resistance in this collection.
This graph is a little bit different, a different way of expressing our total percent resistance by decade. So as you can see, the decades are listed along the bottom but we had, especially for the earlier time points, only a very few numbers strains. So we used normalization of the number of strains just to give you a better idea without skewing for the higher number of strains what the resistance looks like by decade.
So just for example, we saw no resistance in the three strains from 1940. But for the 1950s there were two strains and one strain was resistant to one antimicrobial. So that gave a normalization of about 0.5.
So you can see the resistance phenotype which is number of drugs that were resistant in each strain by the number of strains on the bottom. And you can see an increasing trend as time goes by.
Now this one is really to show you a look at the older drugs as well as some of the newer drugs by decade. So in the blue you see pre-1969 strains. In red we have strains from 1970 to 1989, as well as those from 1990 to 2006 represented. So we have ampicillin resistance, tetracycline, strep, and sulfamethoxazole and the numbers of isolates that we had resistant to each of these drugs.
As you can see for these older drugs, we showed very low resistance, less than 10 in all strains pre-1969. From 1970 to 1989, we saw an increase to around an average of about 28 percent. Then from 1990 to 2006 we saw a dramatic increase in the number of resistant strains.
One particular caveat I wanted to put forward as a limitation of this study was when we asked our partners at CDC to send us strains, we asked them to send us a variety of resistance phenotypes. So when we do write up this project in the very end we will be looking at the overall NARMS data to show more direct prevalence of resistance instead of a bit of a skewed result from asking for particular phenotypes.
But however, I think the take-home message is basically the same. We see much more resistance in later times rather than the earlier times and zero resistance to ceftiofur or ceftriaxone, the third generation cephalosporins pre 1990.
What this is is a dendogram of all 145 strains that were analyzed. The dendogram is based on gene presence or absence called via the microarray.
So what we did, and I know you can’t see any of these numbers, but this gives you an overall picture by decade basically of how these strains clustered by similarity based on all genes that were present or absent.
So, for example, if two strains had more of the genes that were present and absent in that same strain, they were determined as to be more similar. So hopefully that is clear to the audience.
However, some of the interesting things that we saw -- and this is being read from the top down so more similar are on the bottom and more dissimilar are at the top or at the outsides of this dendogram. You can see that we have color coded via the decades. So in particular, turquoise or navy were the 1990 and 2000 strains whereas green would be 1960, and orange for 1970.
So you can see in some cases as we lay down this color coding by decade that some of the strains do cluster by decade. You can see those from 1990 and 2000 are much more dissimilar to any of the other strains from the earlier time points.
We do have some clusters where this does not hold, but I tried to highlight some of these areas where we saw clusters by time.
This is just reiterating what I have just said, basically in more depth than what I just told you. We had a particular cluster of strains from 1965 that clustered with high similarity to each other.
We also had another cluster of strains that was interesting from 1971 that clustered very similarly to each other. Again, those from 1990 -- it is specifically 1999 to 2002 -- clustered together and were most dissimilar from those that were earlier in time.
Also strains from the ‘90s and 2000s rarely clustered with high similarity to those that were from 1960 and below. So you can see the two ends of the spectrum. Those that are earlier in time from 1960 and previous were more dissimilar to those from more recent times which I think is an interesting point.
Secondly, this also included our resistance genes. So it was not just the genomic content but also resistance genes and we saw in some cases where an antimicrobial resistance phenotype and genotype found a particular cluster of similar strains.
So just for example, we found our nalidixic acid-resistant isolates from the 1970s clustering together in an almost indistinguishable cluster, as well as some of the Amp, Chl, and streptomycin resistance isolates clustered together with high similarity. So in some cases we really find where resistance phenotype and genotype are clustering together as well as decades are clustering together, and in other cases where that is not the case.
This is basically just to show you a pretty heat map. You can’t have one of these presentations without a pretty heat map. But what this is showing you is by the axis here are the probe names which represent each gene set that we have included on the array by strains. So you can see how the strains are clustering more similarly to each other by probe sets.
The red indicates that these particular genes are present. Green indicates these particular genes are absent. So by using this sort of a whole genomic approach, although it is not a sequenced approach, we include all of the genes that were included in each of these phenotypes. You can see the clustering of high similarity.
I know you can’t see any of this, but some of these genes that are clustering together in red are genes that you would expect, such as O antigen genes, enterotoxin genes. Some of them are resistance genes, particularly in this case down here. So really this is just to show the beauty of the microarray.
Now this is a bit more of an in-depth perception and I really apologize that I am really giving you sort of the tip of the iceberg of this project. With the number of strains that we are looking at and the number of genes that we are looking at, it is really difficult to give you a really in-depth overview of what we are seeing. But just a little bit of in-depth detection of the resistance genes that we were looking at, we have made this table.
The gray boxes indicate where genotype and phenotype are matching. The way that this table is arranged is by less resistance at the top to greater resistance at the bottom. We tried to throw in there a few isolates from different decades to give you a sort of an appreciation of what we are finding.
We have antimicrobial resistance classes at the top, with some of the more virulent strains that we were looking at, including the pathogenicity islands that Rebecca Lindsey talked about, as well as some of the secretion systems.
So we are finding some interesting things and I was hoping this would be a bit bigger on the screen, but a couple of interesting ones to point out, just to illustrate sort of what I am trying to show you. Here is our SAR-A2 strain was the pansusceptible strain, that is the LT2 sequence strain. We find no resistance in that particular strain and no genes that would have corresponded to resistance. So that was validating what we are seeing.
We also saw some interesting virulence genes such as the O antigen genes, rml. We found some flagellar genes, fliC. We have phage genes that we are finding as well. Some of the virulence, the pathogenicity plasmid, Salmonella virulence plasmid, as well as enterotoxin. A few of the Salmonella pathogenicity island genes and sopD was one of our secretion system genes that we found.
That strain was the LT2, as I said. Just to illustrate one from the 1950’s, the next one down is one from 1957 resistant to tetracycline. We detected tetB and another tet gene that didn’t have an allele assigned to it, as well as some mobile elements.
One of the take-home messages on this particular study that we found is what I think you would expect. So it is really nice to see this data that supports what you would expect and the fact that you are finding more mobile elements -- and this is the mobile element column -- more mobile elements as the increase in resistance phenotype and genotype are being displayed.
So just for example, the very bottom is one of the ones that our compadres from CDC sent us from 2000, resistant to 11 genes and you can see a number of aminoglycoside-resistance genes, or excuse me, resistant to a number of 11 different phenotypes of antimicrobial resistance.
We found aminoglycoside genes, beta-lactamases, chloramphenicol-resistance genes, sulfonamides, tetracycline-resistance genes. Also the sugE gene that we have talked about previously which was published in the paper that our group published with TIGR and some of the folks form CFSAN.
Again, the mercury resistance that Rebecca Lindsey had talked about, along with qacEDelta. Honestly, I think, you know, we are seeing these genes that we would expect to see with class I integrons, with transposons, with plasmids, or even composites of those. So I think really this type of a tool is really great to use to link these things back together.
Obviously, sequencing is great, but I don’t know how feasible it will be until the price really comes down to sequence everything under the sun. So the more we can screen for these things using this type of an array or the other types of arrays that have been talked about earlier, is really going to add to what we know about resistance.
So just in summary, we found 43.5 percent of these strains were susceptible to all antimicrobials and that is from 1948 to 2006. Of the 82 resistant strains, 24 percent were resistant to four or more antimicrobials.
As I illustrated below in a couple of those charts, the resistance was observed to increase dramatically from a low level in 1940s and 1950s to higher levels from 1965 and above. The resistance to our older drugs, as I showed you in that slide which partitioned the bar graph, we found below 10 percent resistance to amp, strep, sulfamethoxazole, and tet in strains pre-1969, average of about 28 percent from 1970 to 1989, and an increase to an average of about 65 percent from 1990 to 2006.
Our resistance to third generation cephalosporins, which as you know are critically important for human health, have occurred post 1990.
Of the resistant strains, 88 percent contained mobile elements such as class 1 integrons, transposase genes, plasmid genes, and the carriage of these mobile elements obviously are very important to our understanding of how these genes are being transferred.
70 percent of these genes were positive for the virulence plasmid, the Salmonella virulence plasmid, as well as the tra locus that Dr. Carattoli was talking about earlier which can indicate the carriage of pSLT which is the very common plasmid that is found in Typhimurium strains.
Strains with multidrug resistant phenotypes were identified exclusively from 1966 to 2006, so we are seeing that multidrug resistance starting in the mid 60s and were positive for multiple mobile elements, most commonly transposase or integrons genes.
100 percent of our resistant isolates were positive for the Salmonella pathogenicity island which was touched upon by Dr. Lindsey, as well as and/or the type III secretion system effector genes.
So really what I sort of want to say at this point is, you know, this has been a collaboration between our two centers, CFSAN and CVM and we have had some back and forth about the particular annotation of the array. I think in some cases we are finding genes with particular alleles that we would expect and other times we are finding them where we don’t expect, which really underlines the points that we either need to validate with sequence strains, which I know my collaborators have done previously, as well as with sequencing our PCR.
One of the things that they have begun to undertake was take all of these probe sets and re-blast them back to NCBI. With one of the ways this array is constructed, similar genes, like for example CMY, when the CMY genes are broken down into the most unique probe set sequence, that CMY allele can encompass 4 to 20 CMY alleles. So really without sequencing or a SNP detection type of tiling, we really would not be able to tell which particular CMY or OXA or PSE that we found. We would not be able to say that with either out sequencing or SNP detection.
So that analysis is currently underway in CFSAN so we will be re-analyzing this data. The lab work is totally finished, so as soon as we do the three analyses. We will be comparing that, of course, back to the AST data which has been our main goal was to look at the rise in phenotype along with the rise in genotype, if you want to call it that.
Some of our staff at DAFM have been doing two-enzyme pulse field on all of these isolates, so we will further then be able to compare particular within the microarray to particular clones within the PFGE and do some really nice comparisons there.
So that will finish the presentation. I just wanted to acknowledge everyone. There have been a lot of people that have done a lot of work on this study, particularly my collaborator at CFSAN, Scott Jackson. Jonathon Sabo has done a lot of the pulse field work and a lot of the DNA hybridizations, as well as our post doc, Dr. Daniel Tadesse, who has done quite a few hybridizations as well. Isha Patel from CFSAN and Sherry Ayres who has done a lot of our re-checking and some of our other AST on some of the other isolates. Of course, my collaborators at CDC, Jason Folster, in particular who sent the DNA and subsequently the strains. Pat McDermott and Shaouha, and our DAFM NARMS team, and the NARMS, CDC team. Of course the FDA Critical Path Initiative.
So come visit us if you get a chance.
DR. FRYE: Thank you Heather, that was great. Our next speaker is going to be Jason Folster with the CDC. He is going to be talking about comparative genomic fingerprinting as a -- excuse me. He is going to be talking about the characterization of blaCMY-encoding plasmids among Salmonella from humans.